diff options
Diffstat (limited to 'generic/tclIO.c')
-rw-r--r-- | generic/tclIO.c | 3178 |
1 files changed, 1247 insertions, 1931 deletions
diff --git a/generic/tclIO.c b/generic/tclIO.c index 715c1ef..d7ed3ac 100644 --- a/generic/tclIO.c +++ b/generic/tclIO.c @@ -6,6 +6,7 @@ * * Copyright (c) 1998-2000 Ajuba Solutions * Copyright (c) 1995-1997 Sun Microsystems, Inc. + * Contributions from Don Porter, NIST, 2014. (not subject to US copyright) * * See the file "license.terms" for information on usage and redistribution of * this file, and for a DISCLAIMER OF ALL WARRANTIES. @@ -16,6 +17,108 @@ #include <assert.h> /* + * For each channel handler registered in a call to Tcl_CreateChannelHandler, + * there is one record of the following type. All of records for a specific + * channel are chained together in a singly linked list which is stored in + * the channel structure. + */ + +typedef struct ChannelHandler { + Channel *chanPtr; /* The channel structure for this channel. */ + int mask; /* Mask of desired events. */ + Tcl_ChannelProc *proc; /* Procedure to call in the type of + * Tcl_CreateChannelHandler. */ + ClientData clientData; /* Argument to pass to procedure. */ + struct ChannelHandler *nextPtr; + /* Next one in list of registered handlers. */ +} ChannelHandler; + +/* + * This structure keeps track of the current ChannelHandler being invoked in + * the current invocation of ChannelHandlerEventProc. There is a potential + * problem if a ChannelHandler is deleted while it is the current one, since + * ChannelHandlerEventProc needs to look at the nextPtr field. To handle this + * problem, structures of the type below indicate the next handler to be + * processed for any (recursively nested) dispatches in progress. The + * nextHandlerPtr field is updated if the handler being pointed to is deleted. + * The nextPtr field is used to chain together all recursive invocations, so + * that Tcl_DeleteChannelHandler can find all the recursively nested + * invocations of ChannelHandlerEventProc and compare the handler being + * deleted against the NEXT handler to be invoked in that invocation; when it + * finds such a situation, Tcl_DeleteChannelHandler updates the nextHandlerPtr + * field of the structure to the next handler. + */ + +typedef struct NextChannelHandler { + ChannelHandler *nextHandlerPtr; /* The next handler to be invoked in + * this invocation. */ + struct NextChannelHandler *nestedHandlerPtr; + /* Next nested invocation of + * ChannelHandlerEventProc. */ +} NextChannelHandler; + +/* + * The following structure describes the event that is added to the Tcl + * event queue by the channel handler check procedure. + */ + +typedef struct ChannelHandlerEvent { + Tcl_Event header; /* Standard header for all events. */ + Channel *chanPtr; /* The channel that is ready. */ + int readyMask; /* Events that have occurred. */ +} ChannelHandlerEvent; + +/* + * The following structure is used by Tcl_GetsObj() to encapsulates the + * state for a "gets" operation. + */ + +typedef struct GetsState { + Tcl_Obj *objPtr; /* The object to which UTF-8 characters + * will be appended. */ + char **dstPtr; /* Pointer into objPtr's string rep where + * next character should be stored. */ + Tcl_Encoding encoding; /* The encoding to use to convert raw bytes + * to UTF-8. */ + ChannelBuffer *bufPtr; /* The current buffer of raw bytes being + * emptied. */ + Tcl_EncodingState state; /* The encoding state just before the last + * external to UTF-8 conversion in + * FilterInputBytes(). */ + int rawRead; /* The number of bytes removed from bufPtr + * in the last call to FilterInputBytes(). */ + int bytesWrote; /* The number of bytes of UTF-8 data + * appended to objPtr during the last call to + * FilterInputBytes(). */ + int charsWrote; /* The corresponding number of UTF-8 + * characters appended to objPtr during the + * last call to FilterInputBytes(). */ + int totalChars; /* The total number of UTF-8 characters + * appended to objPtr so far, just before the + * last call to FilterInputBytes(). */ +} GetsState; + +/* + * The following structure encapsulates the state for a background channel + * copy. Note that the data buffer for the copy will be appended to this + * structure. + */ + +typedef struct CopyState { + struct Channel *readPtr; /* Pointer to input channel. */ + struct Channel *writePtr; /* Pointer to output channel. */ + int readFlags; /* Original read channel flags. */ + int writeFlags; /* Original write channel flags. */ + Tcl_WideInt toRead; /* Number of bytes to copy, or -1. */ + Tcl_WideInt total; /* Total bytes transferred (written). */ + Tcl_Interp *interp; /* Interp that started the copy. */ + Tcl_Obj *cmdPtr; /* Command to be invoked at completion. */ + int bufSize; /* Size of appended buffer. */ + char buffer[1]; /* Copy buffer, this must be the last + * field. */ +} CopyState; + +/* * All static variables used in this file are collected into a single instance * of the following structure. For multi-threaded implementations, there is * one instance of this structure for each thread. @@ -44,15 +147,29 @@ typedef struct ThreadSpecificData { static Tcl_ThreadDataKey dataKey; /* + * Structure to record a close callback. One such record exists for + * each close callback registered for a channel. + */ + +typedef struct CloseCallback { + Tcl_CloseProc *proc; /* The procedure to call. */ + ClientData clientData; /* Arbitrary one-word data to pass + * to the callback. */ + struct CloseCallback *nextPtr; /* For chaining close callbacks. */ +} CloseCallback; + +/* * Static functions in this file: */ static ChannelBuffer * AllocChannelBuffer(int length); +static void PreserveChannelBuffer(ChannelBuffer *bufPtr); +static void ReleaseChannelBuffer(ChannelBuffer *bufPtr); +static int IsShared(ChannelBuffer *bufPtr); static void ChannelTimerProc(ClientData clientData); +static int ChanRead(Channel *chanPtr, char *dst, int dstSize); static int CheckChannelErrors(ChannelState *statePtr, int direction); -static int CheckFlush(Channel *chanPtr, ChannelBuffer *bufPtr, - int newlineFlag); static int CheckForDeadChannel(Tcl_Interp *interp, ChannelState *statePtr); static void CheckForStdChannelsBeingClosed(Tcl_Channel chan); @@ -64,9 +181,6 @@ static int CloseChannelPart(Tcl_Interp *interp, Channel *chanPtr, int errorCode, int flags); static int CloseWrite(Tcl_Interp *interp, Channel *chanPtr); static void CommonGetsCleanup(Channel *chanPtr); -static int CopyAndTranslateBuffer(ChannelState *statePtr, - char *result, int space); -static int CopyBuffer(Channel *chanPtr, char *result, int space); static int CopyData(CopyState *csPtr, int mask); static void CopyEventProc(ClientData clientData, int mask); static void CreateScriptRecord(Tcl_Interp *interp, @@ -79,16 +193,16 @@ static int DetachChannel(Tcl_Interp *interp, Tcl_Channel chan); static void DiscardInputQueued(ChannelState *statePtr, int discardSavedBuffers); static void DiscardOutputQueued(ChannelState *chanPtr); -static int DoRead(Channel *chanPtr, char *srcPtr, int slen, int allowShortReads); -static int DoWrite(Channel *chanPtr, const char *src, int srcLen); +static int DoRead(Channel *chanPtr, char *dst, int bytesToRead, + int allowShortReads); static int DoReadChars(Channel *chan, Tcl_Obj *objPtr, int toRead, int appendFlag); -static int DoWriteChars(Channel *chan, const char *src, int len); static int FilterInputBytes(Channel *chanPtr, GetsState *statePtr); static int FlushChannel(Tcl_Interp *interp, Channel *chanPtr, int calledFromAsyncFlush); static int TclGetsObjBinary(Tcl_Channel chan, Tcl_Obj *objPtr); +static Tcl_Encoding GetBinaryEncoding(); static void FreeBinaryEncoding(ClientData clientData); static Tcl_HashTable * GetChannelTable(Tcl_Interp *interp); static int GetInput(Channel *chanPtr); @@ -97,29 +211,30 @@ static int HaveVersion(const Tcl_ChannelType *typePtr, static void PeekAhead(Channel *chanPtr, char **dstEndPtr, GetsState *gsPtr); static int ReadBytes(ChannelState *statePtr, Tcl_Obj *objPtr, - int charsLeft, int *offsetPtr); + int charsLeft); static int ReadChars(ChannelState *statePtr, Tcl_Obj *objPtr, - int charsLeft, int *offsetPtr, int *factorPtr); + int charsLeft, int *factorPtr); static void RecycleBuffer(ChannelState *statePtr, ChannelBuffer *bufPtr, int mustDiscard); static int StackSetBlockMode(Channel *chanPtr, int mode); static int SetBlockMode(Tcl_Interp *interp, Channel *chanPtr, int mode); static void StopCopy(CopyState *csPtr); -static int TranslateInputEOL(ChannelState *statePtr, char *dst, - const char *src, int *dstLenPtr, int *srcLenPtr); -static int TranslateOutputEOL(ChannelState *statePtr, char *dst, +static void TranslateInputEOL(ChannelState *statePtr, char *dst, const char *src, int *dstLenPtr, int *srcLenPtr); static void UpdateInterest(Channel *chanPtr); -static int WriteBytes(Channel *chanPtr, const char *src, - int srcLen); -static int WriteChars(Channel *chanPtr, const char *src, - int srcLen); +static int Write(Channel *chanPtr, const char *src, + int srcLen, Tcl_Encoding encoding); static Tcl_Obj * FixLevelCode(Tcl_Obj *msg); static void SpliceChannel(Tcl_Channel chan); static void CutChannel(Tcl_Channel chan); static int WillRead(Channel *chanPtr); +#define WriteChars(chanPtr, src, srcLen) \ + Write(chanPtr, src, srcLen, chanPtr->state->encoding) +#define WriteBytes(chanPtr, src, srcLen) \ + Write(chanPtr, src, srcLen, tclIdentityEncoding) + /* * Simplifying helper macros. All may use their argument(s) multiple times. * The ANSI C "prototypes" for the macros are listed below, together with a @@ -172,7 +287,7 @@ static int WillRead(Channel *chanPtr); #define IsBufferEmpty(bufPtr) ((bufPtr)->nextAdded == (bufPtr)->nextRemoved) -#define IsBufferFull(bufPtr) ((bufPtr)->nextAdded >= (bufPtr)->bufLength) +#define IsBufferFull(bufPtr) ((bufPtr) && (bufPtr)->nextAdded >= (bufPtr)->bufLength) #define IsBufferOverflowing(bufPtr) ((bufPtr)->nextAdded>(bufPtr)->bufLength) @@ -209,23 +324,22 @@ static int WillRead(Channel *chanPtr); static void DupChannelIntRep(Tcl_Obj *objPtr, Tcl_Obj *copyPtr); static int SetChannelFromAny(Tcl_Interp *interp, Tcl_Obj *objPtr); -static void UpdateStringOfChannel(Tcl_Obj *objPtr); static void FreeChannelIntRep(Tcl_Obj *objPtr); -static const Tcl_ObjType tclChannelType = { +static const Tcl_ObjType chanObjType = { "channel", /* name for this type */ FreeChannelIntRep, /* freeIntRepProc */ DupChannelIntRep, /* dupIntRepProc */ - NULL, /* updateStringProc UpdateStringOfChannel */ + NULL, /* updateStringProc */ NULL /* setFromAnyProc SetChannelFromAny */ }; #define GET_CHANNELSTATE(objPtr) \ - ((ChannelState *) (objPtr)->internalRep.otherValuePtr) + ((ChannelState *) (objPtr)->internalRep.twoPtrValue.ptr1) #define SET_CHANNELSTATE(objPtr, storePtr) \ - ((objPtr)->internalRep.otherValuePtr = (void *) (storePtr)) + ((objPtr)->internalRep.twoPtrValue.ptr1 = (void *) (storePtr)) #define GET_CHANNELINTERP(objPtr) \ - ((Interp *) (objPtr)->internalRep.twoPtrValue.ptr2) + ((Tcl_Interp *) (objPtr)->internalRep.twoPtrValue.ptr2) #define SET_CHANNELINTERP(objPtr, storePtr) \ ((objPtr)->internalRep.twoPtrValue.ptr2 = (void *) (storePtr)) @@ -268,19 +382,79 @@ ChanCloseHalf( return chanPtr->typePtr->close2Proc(chanPtr->instanceData, interp, flags); } -static inline int +/* + *--------------------------------------------------------------------------- + * + * ChanRead -- + * + * Read up to dstSize bytes using the inputProc of chanPtr, store + * them at dst, and return the number of bytes stored. + * + * Results: + * The return value of the driver inputProc, + * - number of bytes stored at dst, ot + * - -1 on error, with a Posix error code available to the + * caller by calling Tcl_GetErrno(). + * + * Side effects: + * The CHANNEL_BLOCKED and CHANNEL_EOF flags of the channel state are + * set as appropriate. + * On EOF, the inputEncodingFlags are set to perform ending operations + * on decoding. + * TODO - Is this really the right place for that? + * + *--------------------------------------------------------------------------- + */ +static int ChanRead( Channel *chanPtr, char *dst, - int dstSize, - int *errnoPtr) + int dstSize) { + int bytesRead, result; + + /* + * If the caller asked for zero bytes, we'd force the inputProc + * to return zero bytes, and then misinterpret that as EOF. + */ + assert(dstSize > 0); + + /* + * Each read op must set the blocked and eof states anew, not let + * the effect of prior reads leak through. + */ + ResetFlag(chanPtr->state, CHANNEL_BLOCKED | CHANNEL_EOF); if (WillRead(chanPtr) < 0) { return -1; } - return chanPtr->typePtr->inputProc(chanPtr->instanceData, dst, dstSize, - errnoPtr); + bytesRead = chanPtr->typePtr->inputProc(chanPtr->instanceData, + dst, dstSize, &result); + + /* Stop any flag leakage through stacked channel levels */ + ResetFlag(chanPtr->state, CHANNEL_BLOCKED | CHANNEL_EOF); + if (bytesRead > 0) { + /* + * If we get a short read, signal up that we may be BLOCKED. + * We should avoid calling the driver because on some + * platforms we will block in the low level reading code even + * though the channel is set into nonblocking mode. + */ + + if (bytesRead < dstSize) { + SetFlag(chanPtr->state, CHANNEL_BLOCKED); + } + } else if (bytesRead == 0) { + SetFlag(chanPtr->state, CHANNEL_EOF); + chanPtr->state->inputEncodingFlags |= TCL_ENCODING_END; + } else if (bytesRead < 0) { + if ((result == EWOULDBLOCK) || (result == EAGAIN)) { + SetFlag(chanPtr->state, CHANNEL_BLOCKED); + result = EAGAIN; + } + Tcl_SetErrno(result); + } + return bytesRead; } static inline Tcl_WideInt @@ -695,6 +869,8 @@ Tcl_DeleteCloseHandler( if ((cbPtr->proc == proc) && (cbPtr->clientData == clientData)) { if (cbPrevPtr == NULL) { statePtr->closeCbPtr = cbPtr->nextPtr; + } else { + cbPrevPtr->nextPtr = cbPtr->nextPtr; } ckfree(cbPtr); break; @@ -1061,15 +1237,6 @@ Tcl_UnregisterChannel( */ if (statePtr->refCount <= 0) { - /* - * Ensure that if there is another buffer, it gets flushed whether or - * not we are doing a background flush. - */ - - if ((statePtr->curOutPtr != NULL) && - IsBufferReady(statePtr->curOutPtr)) { - SetFlag(statePtr, BUFFER_READY); - } Tcl_Preserve(statePtr); if (!GotFlag(statePtr, BG_FLUSH_SCHEDULED)) { /* @@ -1465,11 +1632,7 @@ Tcl_CreateChannel( statePtr->timer = NULL; statePtr->csPtrR = NULL; statePtr->csPtrW = NULL; - statePtr->outputStage = NULL; - if ((statePtr->encoding != NULL) && GotFlag(statePtr, TCL_WRITABLE)) { - statePtr->outputStage = ckalloc(statePtr->bufSize + 2); - } /* * As we are creating the channel, it is obviously the top for now. @@ -1634,6 +1797,10 @@ Tcl_StackChannel( statePtr->csPtrR = NULL; statePtr->csPtrW = NULL; + /* + * TODO: Examine what can go wrong if Tcl_Flush() call disturbs + * the stacking state of this channel during its operations. + */ if (Tcl_Flush((Tcl_Channel) prevChanPtr) != TCL_OK) { statePtr->csPtrR = csPtrR; statePtr->csPtrW = csPtrW; @@ -1665,17 +1832,17 @@ Tcl_StackChannel( */ if (((mask & TCL_READABLE) != 0) && (statePtr->inQueueHead != NULL)) { + /* - * Remark: It is possible that the channel buffers contain data from - * some earlier push-backs. + * When statePtr->inQueueHead is not NULL, we know + * prevChanPtr->inQueueHead must be NULL. */ - statePtr->inQueueTail->nextPtr = prevChanPtr->inQueueHead; - prevChanPtr->inQueueHead = statePtr->inQueueHead; + assert(prevChanPtr->inQueueHead == NULL); + assert(prevChanPtr->inQueueTail == NULL); - if (prevChanPtr->inQueueTail == NULL) { - prevChanPtr->inQueueTail = statePtr->inQueueTail; - } + prevChanPtr->inQueueHead = statePtr->inQueueHead; + prevChanPtr->inQueueTail = statePtr->inQueueTail; statePtr->inQueueHead = NULL; statePtr->inQueueTail = NULL; @@ -1763,6 +1930,13 @@ Tcl_UnstackChannel( * into the old structure. */ + /* + * TODO: Figure out how to handle the situation where the chan + * operations called below by this unstacking operation cause + * another unstacking recursively. In that case the downChanPtr + * value we're holding on to will not be the right thing. + */ + Channel *downChanPtr = chanPtr->downChanPtr; /* @@ -1867,7 +2041,7 @@ Tcl_UnstackChannel( */ Tcl_EventuallyFree(chanPtr, TCL_DYNAMIC); - UpdateInterest(downChanPtr); + UpdateInterest(statePtr->topChanPtr); if (result != 0) { Tcl_SetErrno(result); @@ -2179,8 +2353,36 @@ AllocChannelBuffer( bufPtr->nextRemoved = BUFFER_PADDING; bufPtr->bufLength = length + BUFFER_PADDING; bufPtr->nextPtr = NULL; + bufPtr->refCount = 1; return bufPtr; } + +static void +PreserveChannelBuffer( + ChannelBuffer *bufPtr) +{ + if (bufPtr->refCount == 0) { + Tcl_Panic("Reuse of ChannelBuffer! %p", bufPtr); + } + bufPtr->refCount++; +} + +static void +ReleaseChannelBuffer( + ChannelBuffer *bufPtr) +{ + if (--bufPtr->refCount) { + return; + } + ckfree(bufPtr); +} + +static int +IsShared( + ChannelBuffer *bufPtr) +{ + return bufPtr->refCount > 1; +} /* *---------------------------------------------------------------------- @@ -2211,20 +2413,23 @@ RecycleBuffer( /* * Do we have to free the buffer to the OS? */ + if (IsShared(bufPtr)) { + mustDiscard = 1; + } if (mustDiscard) { - ckfree(bufPtr); + ReleaseChannelBuffer(bufPtr); return; } /* - * Only save buffers which are at least as big as the requested buffersize - * for the channel. This is to honor dynamic changes of the buffersize + * Only save buffers which have the requested buffersize for the + * channel. This is to honor dynamic changes of the buffersize * made by the user. */ - if ((bufPtr->bufLength - BUFFER_PADDING) < statePtr->bufSize) { - ckfree(bufPtr); + if ((bufPtr->bufLength - BUFFER_PADDING) != statePtr->bufSize) { + ReleaseChannelBuffer(bufPtr); return; } @@ -2259,7 +2464,7 @@ RecycleBuffer( * If we reached this code we return the buffer to the OS. */ - ckfree(bufPtr); + ReleaseChannelBuffer(bufPtr); return; keepBuffer: @@ -2297,6 +2502,11 @@ DiscardOutputQueued( } statePtr->outQueueHead = NULL; statePtr->outQueueTail = NULL; + bufPtr = statePtr->curOutPtr; + if (bufPtr && BytesLeft(bufPtr)) { + statePtr->curOutPtr = NULL; + RecycleBuffer(statePtr, bufPtr, 0); + } } /* @@ -2364,8 +2574,6 @@ FlushChannel( ChannelState *statePtr = chanPtr->state; /* State of the channel stack. */ ChannelBuffer *bufPtr; /* Iterates over buffered output queue. */ - int toWrite; /* Amount of output data in current buffer - * available to be written. */ int written; /* Amount of output data actually written in * current round. */ int errorCode = 0; /* Stores POSIX error codes from channel @@ -2385,63 +2593,58 @@ FlushChannel( } /* - * Loop over the queued buffers and attempt to flush as much as possible - * of the queued output to the channel. + * Should we shift the current output buffer over to the output queue? + * First check that there are bytes in it. If so then... + * If the output queue is empty, then yes, trusting the caller called + * us only when written bytes ought to be flushed. + * If the current output buffer is full, then yes, so we can meet + * the post-condition that on a successful return to caller we've + * left space in the current output buffer for more writing (the flush + * call was to make new room). + * Otherwise, no. Keep the current output buffer where it is so more + * can be written to it, possibly filling it, to promote more efficient + * buffer usage. */ - Tcl_Preserve(chanPtr); - while (1) { - /* - * If the queue is empty and there is a ready current buffer, OR if - * the current buffer is full, then move the current buffer to the - * queue. - */ - - if (((statePtr->curOutPtr != NULL) && - IsBufferFull(statePtr->curOutPtr)) - || (GotFlag(statePtr, BUFFER_READY) && - (statePtr->outQueueHead == NULL))) { - ResetFlag(statePtr, BUFFER_READY); - statePtr->curOutPtr->nextPtr = NULL; - if (statePtr->outQueueHead == NULL) { - statePtr->outQueueHead = statePtr->curOutPtr; - } else { - statePtr->outQueueTail->nextPtr = statePtr->curOutPtr; - } - statePtr->outQueueTail = statePtr->curOutPtr; - statePtr->curOutPtr = NULL; + bufPtr = statePtr->curOutPtr; + if (bufPtr && BytesLeft(bufPtr) && /* Keep empties off queue */ + (statePtr->outQueueHead == NULL || IsBufferFull(bufPtr))) { + if (statePtr->outQueueHead == NULL) { + statePtr->outQueueHead = bufPtr; + } else { + statePtr->outQueueTail->nextPtr = bufPtr; } - bufPtr = statePtr->outQueueHead; + statePtr->outQueueTail = bufPtr; + statePtr->curOutPtr = NULL; + } - /* - * If we are not being called from an async flush and an async flush - * is active, we just return without producing any output. - */ + assert(!IsBufferFull(statePtr->curOutPtr)); - if (!calledFromAsyncFlush && GotFlag(statePtr, BG_FLUSH_SCHEDULED)) { - errorCode = 0; - goto done; - } + /* + * If we are not being called from an async flush and an async flush + * is active, we just return without producing any output. + */ - /* - * If the output queue is still empty, break out of the while loop. - */ + if (!calledFromAsyncFlush && GotFlag(statePtr, BG_FLUSH_SCHEDULED)) { + return 0; + } - if (bufPtr == NULL) { - break; /* Out of the "while (1)". */ - } + /* + * Loop over the queued buffers and attempt to flush as much as possible + * of the queued output to the channel. + */ + + Tcl_Preserve(chanPtr); + while (statePtr->outQueueHead) { + bufPtr = statePtr->outQueueHead; /* * Produce the output on the channel. */ - toWrite = BytesLeft(bufPtr); - if (toWrite == 0) { - written = 0; - } else { - written = ChanWrite(chanPtr, RemovePoint(bufPtr), toWrite, - &errorCode); - } + PreserveChannelBuffer(bufPtr); + written = ChanWrite(chanPtr, RemovePoint(bufPtr), BytesLeft(bufPtr), + &errorCode); /* * If the write failed completely attempt to start the asynchronous @@ -2456,6 +2659,7 @@ FlushChannel( if (errorCode == EINTR) { errorCode = 0; + ReleaseChannelBuffer(bufPtr); continue; } @@ -2477,6 +2681,7 @@ FlushChannel( UpdateInterest(chanPtr); } errorCode = 0; + ReleaseChannelBuffer(bufPtr); break; } @@ -2538,14 +2743,15 @@ FlushChannel( */ DiscardOutputQueued(statePtr); - continue; + ReleaseChannelBuffer(bufPtr); + break; } else { + /* TODO: Consider detecting and reacting to short writes + * on blocking channels. Ought not happen. See iocmd-24.2. */ wroteSome = 1; } - if (!IsBufferEmpty(bufPtr)) { - bufPtr->nextRemoved += written; - } + bufPtr->nextRemoved += written; /* * If this buffer is now empty, recycle it. @@ -2558,7 +2764,8 @@ FlushChannel( } RecycleBuffer(statePtr, bufPtr, 0); } - } /* Closes "while (1)". */ + ReleaseChannelBuffer(bufPtr); + } /* Closes "while". */ /* * If we wrote some data while flushing in the background, we are done. @@ -2573,6 +2780,19 @@ FlushChannel( } else if (statePtr->outQueueHead == NULL) { ResetFlag(statePtr, BG_FLUSH_SCHEDULED); ChanWatch(chanPtr, statePtr->interestMask); + } else { + + /* + * When we are calledFromAsyncFlush, that means a writable + * state on the channel triggered the call, so we should be + * able to write something. Either we did write something + * and wroteSome should be set, or there was nothing left to + * write in this call, and we've completed the BG flush. + * These are the two cases above. If we get here, that means + * there is some kind failure in the writable event machinery. + */ + + assert(!calledFromAsyncFlush); } } @@ -2659,7 +2879,7 @@ CloseChannel( */ if (statePtr->curOutPtr != NULL) { - ckfree(statePtr->curOutPtr); + ReleaseChannelBuffer(statePtr->curOutPtr); statePtr->curOutPtr = NULL; } @@ -2722,10 +2942,6 @@ CloseChannel( } Tcl_FreeEncoding(statePtr->encoding); - if (statePtr->outputStage != NULL) { - ckfree(statePtr->outputStage); - statePtr->outputStage = NULL; - } } /* @@ -3075,10 +3291,18 @@ Tcl_Close( stickyError = 0; - if ((statePtr->encoding != NULL) && (statePtr->curOutPtr != NULL) - && (CheckChannelErrors(statePtr, TCL_WRITABLE) == 0)) { - statePtr->outputEncodingFlags |= TCL_ENCODING_END; - if (WriteChars(chanPtr, "", 0) < 0) { + if (GotFlag(statePtr, TCL_WRITABLE) && (statePtr->encoding != NULL) + && !(statePtr->outputEncodingFlags & TCL_ENCODING_START)) { + + int code = CheckChannelErrors(statePtr, TCL_WRITABLE); + + if (code == 0) { + statePtr->outputEncodingFlags |= TCL_ENCODING_END; + code = WriteChars(chanPtr, "", 0); + statePtr->outputEncodingFlags &= ~TCL_ENCODING_END; + statePtr->outputEncodingFlags |= TCL_ENCODING_START; + } + if (code < 0) { stickyError = Tcl_GetErrno(); } @@ -3113,14 +3337,6 @@ Tcl_Close( ResetFlag(statePtr, CHANNEL_INCLOSE); /* - * Ensure that the last output buffer will be flushed. - */ - - if ((statePtr->curOutPtr != NULL) && IsBufferReady(statePtr->curOutPtr)) { - SetFlag(statePtr, BUFFER_READY); - } - - /* * If this channel supports it, close the read side, since we don't need * it anymore and this will help avoid deadlocks on some channel types. */ @@ -3165,7 +3381,17 @@ Tcl_Close( Tcl_SetObjResult(interp, Tcl_NewStringObj(Tcl_PosixError(interp), -1)); } - flushcode = -1; + return TCL_ERROR; + } + /* + * Bug 97069ea11a: set error message if a flush code is set and no error + * message set up to now. + */ + if (flushcode != 0 && interp != NULL + && 0 == Tcl_GetCharLength(Tcl_GetObjResult(interp)) ) { + Tcl_SetErrno(flushcode); + Tcl_SetObjResult(interp, + Tcl_NewStringObj(Tcl_PosixError(interp), -1)); } if ((flushcode != 0) || (result != 0)) { return TCL_ERROR; @@ -3278,10 +3504,6 @@ Tcl_CloseEx( return CloseChannelPart(interp, chanPtr, 0, flags); } else if (flags & TCL_CLOSE_WRITE) { - if ((statePtr->curOutPtr != NULL) && - IsBufferReady(statePtr->curOutPtr)) { - SetFlag(statePtr, BUFFER_READY); - } Tcl_Preserve(statePtr); if (!GotFlag(statePtr, BG_FLUSH_SCHEDULED)) { /* @@ -3343,14 +3565,6 @@ CloseWrite( int result = 0; /* - * Ensure that the last output buffer will be flushed. - */ - - if ((statePtr->curOutPtr != NULL) && IsBufferReady(statePtr->curOutPtr)) { - SetFlag(statePtr, BUFFER_READY); - } - - /* * The call to FlushChannel will flush any queued output and invoke the * close function of the channel driver, or it will set up the channel to * be flushed and closed asynchronously. @@ -3662,7 +3876,7 @@ Tcl_Write( if (srcLen < 0) { srcLen = strlen(src); } - return DoWrite(chanPtr, src, srcLen); + return WriteBytes(chanPtr, src, srcLen); } /* @@ -3753,82 +3967,40 @@ Tcl_WriteChars( int len) /* Length of string in bytes, or < 0 for * strlen(). */ { - ChannelState *statePtr; /* State info for channel */ - - statePtr = ((Channel *) chan)->state; + Channel *chanPtr = (Channel *) chan; + ChannelState *statePtr = chanPtr->state; /* State info for channel */ + int result; + Tcl_Obj *objPtr; if (CheckChannelErrors(statePtr, TCL_WRITABLE) != 0) { return -1; } - return DoWriteChars((Channel *) chan, src, len); -} - -/* - *--------------------------------------------------------------------------- - * - * DoWriteChars -- - * - * Takes a sequence of UTF-8 characters and converts them for output - * using the channel's current encoding, may queue the buffer for output - * if it gets full, and also remembers whether the current buffer is - * ready e.g. if it contains a newline and we are in line buffering mode. - * Compensates stacking, i.e. will redirect the data from the specified - * channel to the topmost channel in a stack. - * - * Results: - * The number of bytes written or -1 in case of error. If -1, - * Tcl_GetErrno will return the error code. - * - * Side effects: - * May buffer up output and may cause output to be produced on the - * channel. - * - *---------------------------------------------------------------------- - */ - -static int -DoWriteChars( - Channel *chanPtr, /* The channel to buffer output for. */ - const char *src, /* UTF-8 characters to queue in output - * buffer. */ - int len) /* Length of string in bytes, or < 0 for - * strlen(). */ -{ - /* - * Always use the topmost channel of the stack - */ - - ChannelState *statePtr; /* State info for channel */ - - statePtr = chanPtr->state; chanPtr = statePtr->topChanPtr; if (len < 0) { len = strlen(src); } - if (statePtr->encoding == NULL) { - /* - * Inefficient way to convert UTF-8 to byte-array, but the code - * parallels the way it is done for objects. - * Special case for 1-byte (used by eg [puts] for the \n) could - * be extended to more efficient translation of the src string. - */ - - int result; + if (statePtr->encoding) { + return WriteChars(chanPtr, src, len); + } - if ((len == 1) && (UCHAR(*src) < 0xC0)) { - result = WriteBytes(chanPtr, src, len); - } else { - Tcl_Obj *objPtr = Tcl_NewStringObj(src, len); + /* + * Inefficient way to convert UTF-8 to byte-array, but the code + * parallels the way it is done for objects. Special case for 1-byte + * (used by eg [puts] for the \n) could be extended to more efficient + * translation of the src string. + */ - src = (char *) Tcl_GetByteArrayFromObj(objPtr, &len); - result = WriteBytes(chanPtr, src, len); - TclDecrRefCount(objPtr); - } - return result; + if ((len == 1) && (UCHAR(*src) < 0xC0)) { + return WriteBytes(chanPtr, src, len); } - return WriteChars(chanPtr, src, len); + + objPtr = Tcl_NewStringObj(src, len); + src = (char *) Tcl_GetByteArrayFromObj(objPtr, &len); + result = WriteBytes(chanPtr, src, len); + TclDecrRefCount(objPtr); + return result; } /* @@ -3904,12 +4076,25 @@ static int WillRead( Channel *chanPtr) { + if (chanPtr->typePtr == NULL) { + /* Prevent read attempts on a closed channel */ + DiscardInputQueued(chanPtr->state, 0); + Tcl_SetErrno(EINVAL); + return -1; + } if ((chanPtr->typePtr->seekProc != NULL) && (Tcl_OutputBuffered((Tcl_Channel) chanPtr) > 0)) { - if ((chanPtr->state->curOutPtr != NULL) - && IsBufferReady(chanPtr->state->curOutPtr)) { - SetFlag(chanPtr->state, BUFFER_READY); - } + + /* + * CAVEAT - The assumption here is that FlushChannel() will + * push out the bytes of any writes that are in progress. + * Since this is a seekable channel, we assume it is not one + * that can block and force bg flushing. Channels we know that + * can do that -- sockets, pipes -- are not seekable. If the + * assumption is wrong, more drastic measures may be required here + * like temporarily setting the channel into blocking mode. + */ + if (FlushChannel(NULL, chanPtr, 0) != 0) { return -1; } @@ -3920,110 +4105,9 @@ WillRead( /* *---------------------------------------------------------------------- * - * WriteBytes -- - * - * Write a sequence of bytes into an output buffer, may queue the buffer - * for output if it gets full, and also remembers whether the current - * buffer is ready e.g. if it contains a newline and we are in line - * buffering mode. - * - * Results: - * The number of bytes written or -1 in case of error. If -1, - * Tcl_GetErrno will return the error code. - * - * Side effects: - * May buffer up output and may cause output to be produced on the - * channel. - * - *---------------------------------------------------------------------- - */ - -static int -WriteBytes( - Channel *chanPtr, /* The channel to buffer output for. */ - const char *src, /* Bytes to write. */ - int srcLen) /* Number of bytes to write. */ -{ - ChannelState *statePtr = chanPtr->state; - /* State info for channel */ - ChannelBuffer *bufPtr; - char *dst; - int dstMax, sawLF, savedLF, total, dstLen, toWrite, translate; - - if (srcLen) { - WillWrite(chanPtr); - } - - total = 0; - sawLF = 0; - savedLF = 0; - translate = GotFlag(statePtr, CHANNEL_LINEBUFFERED) - || (statePtr->outputTranslation != TCL_TRANSLATE_LF); - - /* - * Loop over all bytes in src, storing them in output buffer with proper - * EOL translation. - */ - - while (srcLen + savedLF > 0) { - bufPtr = statePtr->curOutPtr; - if (bufPtr == NULL) { - bufPtr = AllocChannelBuffer(statePtr->bufSize); - statePtr->curOutPtr = bufPtr; - } - dst = InsertPoint(bufPtr); - dstMax = SpaceLeft(bufPtr); - dstLen = dstMax; - - toWrite = dstLen; - if (toWrite > srcLen) { - toWrite = srcLen; - } - - if (translate) { - if (savedLF) { - /* - * A '\n' was left over from last call to TranslateOutputEOL() - * and we need to store it in this buffer. If the channel is - * line-based, we will need to flush it. - */ - - *dst++ = '\n'; - dstLen--; - sawLF++; - } - if (TranslateOutputEOL(statePtr, dst, src, &dstLen, &toWrite)) { - sawLF++; - } - dstLen += savedLF; - savedLF = 0; - if (dstLen > dstMax) { - savedLF = 1; - dstLen = dstMax; - } - } else { - memcpy(dst, src, toWrite); - dstLen = toWrite; - } - - bufPtr->nextAdded += dstLen; - if (CheckFlush(chanPtr, bufPtr, sawLF) != 0) { - return -1; - } - total += dstLen; - src += toWrite; - srcLen -= toWrite; - sawLF = 0; - } - return total; -} - -/* - *---------------------------------------------------------------------- - * - * WriteChars -- + * Write -- * - * Convert UTF-8 bytes to the channel's external encoding and write the + * Convert srcLen bytes starting at src according to encoding and write * produced bytes into an output buffer, may queue the buffer for output * if it gets full, and also remembers whether the current buffer is * ready e.g. if it contains a newline and we are in line buffering mode. @@ -4040,381 +4124,164 @@ WriteBytes( */ static int -WriteChars( +Write( Channel *chanPtr, /* The channel to buffer output for. */ const char *src, /* UTF-8 string to write. */ - int srcLen) /* Length of UTF-8 string in bytes. */ + int srcLen, /* Length of UTF-8 string in bytes. */ + Tcl_Encoding encoding) { ChannelState *statePtr = chanPtr->state; /* State info for channel */ - ChannelBuffer *bufPtr; - char *dst, *stage; - int saved, savedLF, sawLF, total, dstLen, stageMax, dstWrote; - int stageLen, toWrite, stageRead, endEncoding, result; - int consumedSomething, translate; - Tcl_Encoding encoding; - char safe[BUFFER_PADDING]; + char *nextNewLine = NULL; + int endEncoding, saved = 0, total = 0, flushed = 0, needNlFlush = 0; if (srcLen) { WillWrite(chanPtr); } - total = 0; - sawLF = 0; - savedLF = 0; - saved = 0; - encoding = statePtr->encoding; - /* * Write the terminated escape sequence even if srcLen is 0. */ endEncoding = ((statePtr->outputEncodingFlags & TCL_ENCODING_END) != 0); - translate = GotFlag(statePtr, CHANNEL_LINEBUFFERED) - || (statePtr->outputTranslation != TCL_TRANSLATE_LF); - - /* - * Loop over all UTF-8 characters in src, storing them in staging buffer - * with proper EOL translation. - */ + if (GotFlag(statePtr, CHANNEL_LINEBUFFERED) + || (statePtr->outputTranslation != TCL_TRANSLATE_LF)) { + nextNewLine = memchr(src, '\n', srcLen); + } - consumedSomething = 1; - while (consumedSomething && (srcLen + savedLF + endEncoding > 0)) { - consumedSomething = 0; - stage = statePtr->outputStage; - stageMax = statePtr->bufSize; - stageLen = stageMax; + while (srcLen + saved + endEncoding > 0) { + ChannelBuffer *bufPtr; + char *dst, safe[BUFFER_PADDING]; + int result, srcRead, dstLen, dstWrote, srcLimit = srcLen; - toWrite = stageLen; - if (toWrite > srcLen) { - toWrite = srcLen; + if (nextNewLine) { + srcLimit = nextNewLine - src; } - - if (translate) { - if (savedLF) { - /* - * A '\n' was left over from last call to TranslateOutputEOL() - * and we need to store it in the staging buffer. If the - * channel is line-based, we will need to flush the output - * buffer (after translating the staging buffer). - */ - - *stage++ = '\n'; - stageLen--; - sawLF++; - } - if (TranslateOutputEOL(statePtr, stage, src, &stageLen, - &toWrite)) { - sawLF++; - } - - stage -= savedLF; - stageLen += savedLF; - savedLF = 0; - - if (stageLen > stageMax) { - savedLF = 1; - stageLen = stageMax; - } - } else { - memcpy(stage, src, toWrite); - stageLen = toWrite; + + /* Get space to write into */ + bufPtr = statePtr->curOutPtr; + if (bufPtr == NULL) { + bufPtr = AllocChannelBuffer(statePtr->bufSize); + statePtr->curOutPtr = bufPtr; } - src += toWrite; - srcLen -= toWrite; - - /* - * Loop over all UTF-8 characters in staging buffer, converting them - * to external encoding, storing them in output buffer. - */ - - while (stageLen + saved + endEncoding > 0) { - bufPtr = statePtr->curOutPtr; - if (bufPtr == NULL) { - bufPtr = AllocChannelBuffer(statePtr->bufSize); - statePtr->curOutPtr = bufPtr; - } - dst = InsertPoint(bufPtr); - dstLen = SpaceLeft(bufPtr); - - if (saved != 0) { - /* - * Here's some translated bytes left over from the last buffer - * that we need to stick at the beginning of this buffer. - */ - - memcpy(dst, safe, (size_t) saved); - bufPtr->nextAdded += saved; - dst += saved; - dstLen -= saved; - saved = 0; - } - - result = Tcl_UtfToExternal(NULL, encoding, stage, stageLen, - statePtr->outputEncodingFlags, - &statePtr->outputEncodingState, dst, - dstLen + BUFFER_PADDING, &stageRead, &dstWrote, NULL); - + if (saved) { /* - * Fix for SF #506297, reported by Martin Forssen - * <ruric@users.sourceforge.net>. - * - * The encoding chosen in the script exposing the bug writes out - * three intro characters when TCL_ENCODING_START is set, but does - * not consume any input as TCL_ENCODING_END is cleared. As some - * output was generated the enclosing loop calls UtfToExternal - * again, again with START set. Three more characters in the out - * and still no use of input ... To break this infinite loop we - * remove TCL_ENCODING_START from the set of flags after the first - * call (no condition is required, the later calls remove an unset - * flag, which is a no-op). This causes the subsequent calls to - * UtfToExternal to consume and convert the actual input. + * Here's some translated bytes left over from the last buffer + * that we need to stick at the beginning of this buffer. */ - statePtr->outputEncodingFlags &= ~TCL_ENCODING_START; + memcpy(InsertPoint(bufPtr), safe, (size_t) saved); + bufPtr->nextAdded += saved; + saved = 0; + } + PreserveChannelBuffer(bufPtr); + dst = InsertPoint(bufPtr); + dstLen = SpaceLeft(bufPtr); + + result = Tcl_UtfToExternal(NULL, encoding, src, srcLimit, + statePtr->outputEncodingFlags, + &statePtr->outputEncodingState, dst, + dstLen + BUFFER_PADDING, &srcRead, &dstWrote, NULL); + + /* See chan-io-1.[89]. Tcl Bug 506297. */ + statePtr->outputEncodingFlags &= ~TCL_ENCODING_START; + + if ((result != TCL_OK) && (srcRead + dstWrote == 0)) { + /* We're reading from invalid/incomplete UTF-8 */ + ReleaseChannelBuffer(bufPtr); + if (total == 0) { + Tcl_SetErrno(EINVAL); + return -1; + } + break; + } - /* - * The following code must be executed only when result is not 0. - */ + bufPtr->nextAdded += dstWrote; + src += srcRead; + srcLen -= srcRead; + total += dstWrote; + dst += dstWrote; + dstLen -= dstWrote; - if ((result != 0) && (stageRead + dstWrote == 0)) { - /* - * We have an incomplete UTF-8 character at the end of the - * staging buffer. It will get moved to the beginning of the - * staging buffer followed by more bytes from src. - */ + if (src == nextNewLine && dstLen > 0) { + static char crln[3] = "\r\n"; + char *nl = NULL; + int nlLen = 0; - src -= stageLen; - srcLen += stageLen; - stageLen = 0; - savedLF = 0; + switch (statePtr->outputTranslation) { + case TCL_TRANSLATE_LF: + nl = crln + 1; + nlLen = 1; + break; + case TCL_TRANSLATE_CR: + nl = crln; + nlLen = 1; + break; + case TCL_TRANSLATE_CRLF: + nl = crln; + nlLen = 2; + break; + default: + Tcl_Panic("unknown output translation requested"); break; } - bufPtr->nextAdded += dstWrote; - if (IsBufferOverflowing(bufPtr)) { - /* - * When translating from UTF-8 to external encoding, we - * allowed the translation to produce a character that crossed - * the end of the output buffer, so that we would get a - * completely full buffer before flushing it. The extra bytes - * will be moved to the beginning of the next buffer. - */ + + result |= Tcl_UtfToExternal(NULL, encoding, nl, nlLen, + statePtr->outputEncodingFlags, + &statePtr->outputEncodingState, dst, + dstLen + BUFFER_PADDING, &srcRead, &dstWrote, NULL); - saved = -SpaceLeft(bufPtr); - memcpy(safe, dst + dstLen, (size_t) saved); - bufPtr->nextAdded = bufPtr->bufLength; - } - if (CheckFlush(chanPtr, bufPtr, sawLF) != 0) { - return -1; - } + assert (srcRead == nlLen); + bufPtr->nextAdded += dstWrote; + src++; + srcLen--; total += dstWrote; - stage += stageRead; - stageLen -= stageRead; - sawLF = 0; - - consumedSomething = 1; + dst += dstWrote; + dstLen -= dstWrote; + nextNewLine = memchr(src, '\n', srcLen); + needNlFlush = 1; + } + if (IsBufferOverflowing(bufPtr)) { /* - * If all translated characters are written to the buffer, - * endEncoding is set to 0 because the escape sequence may be - * output. + * When translating from UTF-8 to external encoding, we + * allowed the translation to produce a character that crossed + * the end of the output buffer, so that we would get a + * completely full buffer before flushing it. The extra bytes + * will be moved to the beginning of the next buffer. */ - if ((stageLen + saved == 0) && (result == 0)) { - endEncoding = 0; - } + saved = -SpaceLeft(bufPtr); + memcpy(safe, dst + dstLen, (size_t) saved); + bufPtr->nextAdded = bufPtr->bufLength; } - } - - /* - * If nothing was written and it happened because there was no progress in - * the UTF conversion, we throw an error. - */ - - if (!consumedSomething && (total == 0)) { - Tcl_SetErrno(EINVAL); - return -1; - } - return total; -} - -/* - *--------------------------------------------------------------------------- - * - * TranslateOutputEOL -- - * - * Helper function for WriteBytes() and WriteChars(). Converts the '\n' - * characters in the source buffer into the appropriate EOL form - * specified by the output translation mode. - * - * EOL translation stops either when the source buffer is empty or the - * output buffer is full. - * - * When converting to CRLF mode and there is only 1 byte left in the - * output buffer, this routine stores the '\r' in the last byte and then - * stores the '\n' in the byte just past the end of the buffer. The - * caller is responsible for passing in a buffer that is large enough to - * hold the extra byte. - * - * Results: - * The return value is 1 if a '\n' was translated from the source buffer, - * or 0 otherwise -- this can be used by the caller to decide to flush a - * line-based channel even though the channel buffer is not full. - * - * *dstLenPtr is filled with how many bytes of the output buffer were - * used. As mentioned above, this can be one more that the output - * buffer's specified length if a CRLF was stored. - * - * *srcLenPtr is filled with how many bytes of the source buffer were - * consumed. - * - * Side effects: - * It may be obvious, but bears mentioning that when converting in CRLF - * mode (which requires two bytes of storage in the output buffer), the - * number of bytes consumed from the source buffer will be less than the - * number of bytes stored in the output buffer. - * - *--------------------------------------------------------------------------- - */ -static int -TranslateOutputEOL( - ChannelState *statePtr, /* Channel being read, for translation and - * buffering modes. */ - char *dst, /* Output buffer filled with UTF-8 chars by - * applying appropriate EOL translation to - * source characters. */ - const char *src, /* Source UTF-8 characters. */ - int *dstLenPtr, /* On entry, the maximum length of output - * buffer in bytes. On exit, the number of - * bytes actually used in output buffer. */ - int *srcLenPtr) /* On entry, the length of source buffer. On - * exit, the number of bytes read from the - * source buffer. */ -{ - char *dstEnd; - int srcLen, newlineFound; - - newlineFound = 0; - srcLen = *srcLenPtr; - - switch (statePtr->outputTranslation) { - case TCL_TRANSLATE_LF: - for (dstEnd = dst + srcLen; dst < dstEnd; ) { - if (*src == '\n') { - newlineFound = 1; - } - *dst++ = *src++; + if ((srcLen + saved == 0) && (result == TCL_OK)) { + endEncoding = 0; } - *dstLenPtr = srcLen; - break; - case TCL_TRANSLATE_CR: - for (dstEnd = dst + srcLen; dst < dstEnd;) { - if (*src == '\n') { - *dst++ = '\r'; - newlineFound = 1; - src++; - } else { - *dst++ = *src++; - } - } - *dstLenPtr = srcLen; - break; - case TCL_TRANSLATE_CRLF: { - /* - * Since this causes the number of bytes to grow, we start off trying - * to put 'srcLen' bytes into the output buffer, but allow it to store - * more bytes, as long as there's still source bytes and room in the - * output buffer. - */ - - char *dstStart, *dstMax; - const char *srcStart; - - dstStart = dst; - dstMax = dst + *dstLenPtr; - - srcStart = src; - if (srcLen < *dstLenPtr) { - dstEnd = dst + srcLen; - } else { - dstEnd = dst + *dstLenPtr; - } - while (dst < dstEnd) { - if (*src == '\n') { - if (dstEnd < dstMax) { - dstEnd++; - } - *dst++ = '\r'; - newlineFound = 1; - } - *dst++ = *src++; - } - *srcLenPtr = src - srcStart; - *dstLenPtr = dst - dstStart; - break; - } - default: - break; - } - return newlineFound; -} - -/* - *--------------------------------------------------------------------------- - * - * CheckFlush -- - * - * Helper function for WriteBytes() and WriteChars(). If the channel - * buffer is ready to be flushed, flush it. - * - * Results: - * The return value is -1 if there was a problem flushing the channel - * buffer, or 0 otherwise. - * - * Side effects: - * The buffer will be recycled if it is flushed. - * - *--------------------------------------------------------------------------- - */ - -static int -CheckFlush( - Channel *chanPtr, /* Channel being read, for buffering mode. */ - ChannelBuffer *bufPtr, /* Channel buffer to possibly flush. */ - int newlineFlag) /* Non-zero if a the channel buffer contains a - * newline. */ -{ - ChannelState *statePtr = chanPtr->state; - /* State info for channel */ - - /* - * The current buffer is ready for output: - * 1. if it is full. - * 2. if it contains a newline and this channel is line-buffered. - * 3. if it contains any output and this channel is unbuffered. - */ - - if (!GotFlag(statePtr, BUFFER_READY)) { if (IsBufferFull(bufPtr)) { - SetFlag(statePtr, BUFFER_READY); - } else if (GotFlag(statePtr, CHANNEL_LINEBUFFERED)) { - if (newlineFlag != 0) { - SetFlag(statePtr, BUFFER_READY); + if (FlushChannel(NULL, chanPtr, 0) != 0) { + ReleaseChannelBuffer(bufPtr); + return -1; + } + flushed += statePtr->bufSize; + if (saved == 0 || src[-1] != '\n') { + needNlFlush = 0; } - } else if (GotFlag(statePtr, CHANNEL_UNBUFFERED)) { - SetFlag(statePtr, BUFFER_READY); } + ReleaseChannelBuffer(bufPtr); } - if (GotFlag(statePtr, BUFFER_READY)) { + if ((flushed < total) && (GotFlag(statePtr, CHANNEL_UNBUFFERED) || + (needNlFlush && GotFlag(statePtr, CHANNEL_LINEBUFFERED)))) { if (FlushChannel(NULL, chanPtr, 0) != 0) { return -1; } } - return 0; + + return total; } /* @@ -4496,8 +4363,7 @@ Tcl_GetsObj( Tcl_EncodingState oldState; if (CheckChannelErrors(statePtr, TCL_READABLE) != 0) { - copiedTotal = -1; - goto done; + return -1; } /* @@ -4517,6 +4383,7 @@ Tcl_GetsObj( */ chanPtr = statePtr->topChanPtr; + Tcl_Preserve(chanPtr); bufPtr = statePtr->inQueueHead; encoding = statePtr->encoding; @@ -4540,16 +4407,7 @@ Tcl_GetsObj( */ if (encoding == NULL) { - ThreadSpecificData *tsdPtr = TCL_TSD_INIT(&dataKey); - - if (tsdPtr->binaryEncoding == NULL) { - tsdPtr->binaryEncoding = Tcl_GetEncoding(NULL, "iso8859-1"); - Tcl_CreateThreadExitHandler(FreeBinaryEncoding, NULL); - } - encoding = tsdPtr->binaryEncoding; - if (encoding == NULL) { - Tcl_Panic("attempted gets on binary channel where no iso8859-1 encoding available"); - } + encoding = GetBinaryEncoding(); } /* @@ -4738,6 +4596,7 @@ Tcl_GetsObj( Tcl_SetObjLength(objPtr, oldLength); CommonGetsCleanup(chanPtr); copiedTotal = -1; + ResetFlag(statePtr, CHANNEL_BLOCKED); goto done; } goto gotEOL; @@ -4759,7 +4618,11 @@ Tcl_GetsObj( * self-modifying reflected transforms. */ - chanPtr = statePtr->topChanPtr; + if (chanPtr != statePtr->topChanPtr) { + Tcl_Release(chanPtr); + chanPtr = statePtr->topChanPtr; + Tcl_Preserve(chanPtr); + } bufPtr = gs.bufPtr; if (bufPtr == NULL) { @@ -4793,16 +4656,18 @@ Tcl_GetsObj( * Regenerate the top channel, in case it was changed due to * self-modifying reflected transforms. */ - - chanPtr = statePtr->topChanPtr; - + if (chanPtr != statePtr->topChanPtr) { + Tcl_Release(chanPtr); + chanPtr = statePtr->topChanPtr; + Tcl_Preserve(chanPtr); + } bufPtr = statePtr->inQueueHead; - if (bufPtr == NULL) { - Tcl_Panic("Tcl_GetsObj: restore reached with bufPtr==NULL"); + if (bufPtr != NULL) { + bufPtr->nextRemoved = oldRemoved; + bufPtr = bufPtr->nextPtr; } - bufPtr->nextRemoved = oldRemoved; - for (bufPtr = bufPtr->nextPtr; bufPtr != NULL; bufPtr = bufPtr->nextPtr) { + for ( ; bufPtr != NULL; bufPtr = bufPtr->nextPtr) { bufPtr->nextRemoved = BUFFER_PADDING; } CommonGetsCleanup(chanPtr); @@ -4835,10 +4700,13 @@ Tcl_GetsObj( * Regenerate the top channel, in case it was changed due to * self-modifying reflected transforms. */ - - chanPtr = statePtr->topChanPtr; - + if (chanPtr != statePtr->topChanPtr) { + Tcl_Release(chanPtr); + chanPtr = statePtr->topChanPtr; + Tcl_Preserve(chanPtr); + } UpdateInterest(chanPtr); + Tcl_Release(chanPtr); return copiedTotal; } @@ -4884,6 +4752,7 @@ TclGetsObjBinary( */ chanPtr = statePtr->topChanPtr; + Tcl_Preserve(chanPtr); bufPtr = statePtr->inQueueHead; @@ -4930,17 +4799,13 @@ TclGetsObjBinary( * hasn't seen EOL. Need to read more bytes from the channel * device. Side effect is to allocate another channel buffer. */ - - if (GotFlag(statePtr, CHANNEL_BLOCKED)) { - if (GotFlag(statePtr, CHANNEL_NONBLOCKING)) { - goto restore; - } - ResetFlag(statePtr, CHANNEL_BLOCKED); - } if (GetInput(chanPtr) != 0) { goto restore; } bufPtr = statePtr->inQueueTail; + if (bufPtr == NULL) { + goto restore; + } } dst = (unsigned char *) RemovePoint(bufPtr); @@ -4995,10 +4860,15 @@ TclGetsObjBinary( byteArray = Tcl_SetByteArrayLength(objPtr, oldLength); CommonGetsCleanup(chanPtr); copiedTotal = -1; + ResetFlag(statePtr, CHANNEL_BLOCKED); goto done; } goto gotEOL; } + if (GotFlag(statePtr, CHANNEL_BLOCKED|CHANNEL_NONBLOCKING) + == (CHANNEL_BLOCKED|CHANNEL_NONBLOCKING)) { + goto restore; + } /* * Copy bytes from the channel buffer to the ByteArray. @@ -5053,12 +4923,12 @@ TclGetsObjBinary( restore: bufPtr = statePtr->inQueueHead; - if (bufPtr == NULL) { - Tcl_Panic("TclGetsObjBinary: restore reached with bufPtr==NULL"); + if (bufPtr) { + bufPtr->nextRemoved = oldRemoved; + bufPtr = bufPtr->nextPtr; } - bufPtr->nextRemoved = oldRemoved; - for (bufPtr = bufPtr->nextPtr; bufPtr != NULL; bufPtr = bufPtr->nextPtr) { + for ( ; bufPtr != NULL; bufPtr = bufPtr->nextPtr) { bufPtr->nextRemoved = BUFFER_PADDING; } CommonGetsCleanup(chanPtr); @@ -5087,6 +4957,7 @@ TclGetsObjBinary( done: UpdateInterest(chanPtr); + Tcl_Release(chanPtr); return copiedTotal; } @@ -5115,6 +4986,21 @@ FreeBinaryEncoding( tsdPtr->binaryEncoding = NULL; } } + +static Tcl_Encoding +GetBinaryEncoding() +{ + ThreadSpecificData *tsdPtr = TCL_TSD_INIT(&dataKey); + + if (tsdPtr->binaryEncoding == NULL) { + tsdPtr->binaryEncoding = Tcl_GetEncoding(NULL, "iso8859-1"); + Tcl_CreateThreadExitHandler(FreeBinaryEncoding, NULL); + } + if (tsdPtr->binaryEncoding == NULL) { + Tcl_Panic("binary encoding is not available"); + } + return tsdPtr->binaryEncoding; +} /* *--------------------------------------------------------------------------- @@ -5148,7 +5034,7 @@ FilterInputBytes( ChannelState *statePtr = chanPtr->state; /* State info for channel */ ChannelBuffer *bufPtr; - char *raw, *rawStart, *dst; + char *raw, *dst; int offset, toRead, dstNeeded, spaceLeft, result, rawLen; Tcl_Obj *objPtr; #define ENCODING_LINESIZE 20 /* Lower bound on how many bytes to convert at @@ -5182,14 +5068,6 @@ FilterInputBytes( */ read: - if (GotFlag(statePtr, CHANNEL_BLOCKED)) { - if (GotFlag(statePtr, CHANNEL_NONBLOCKING)) { - gsPtr->charsWrote = 0; - gsPtr->rawRead = 0; - return -1; - } - ResetFlag(statePtr, CHANNEL_BLOCKED); - } if (GetInput(chanPtr) != 0) { gsPtr->charsWrote = 0; gsPtr->rawRead = 0; @@ -5197,6 +5075,11 @@ FilterInputBytes( } bufPtr = statePtr->inQueueTail; gsPtr->bufPtr = bufPtr; + if (bufPtr == NULL) { + gsPtr->charsWrote = 0; + gsPtr->rawRead = 0; + return -1; + } } /* @@ -5205,8 +5088,7 @@ FilterInputBytes( * string rep if we need more space. */ - rawStart = RemovePoint(bufPtr); - raw = rawStart; + raw = RemovePoint(bufPtr); rawLen = BytesLeft(bufPtr); dst = *gsPtr->dstPtr; @@ -5273,9 +5155,15 @@ FilterInputBytes( } else { /* * There are no more cached raw bytes left. See if we can get - * some more. + * some more, but avoid blocking on a non-blocking channel. */ + if (GotFlag(statePtr, CHANNEL_NONBLOCKING|CHANNEL_BLOCKED) + == (CHANNEL_NONBLOCKING|CHANNEL_BLOCKED)) { + gsPtr->charsWrote = 0; + gsPtr->rawRead = 0; + return -1; + } goto read; } } else { @@ -5514,134 +5402,77 @@ Tcl_Read( int Tcl_ReadRaw( Tcl_Channel chan, /* The channel from which to read. */ - char *bufPtr, /* Where to store input read. */ + char *readBuf, /* Where to store input read. */ int bytesToRead) /* Maximum number of bytes to read. */ { Channel *chanPtr = (Channel *) chan; ChannelState *statePtr = chanPtr->state; /* State info for channel */ - int nread, result, copied, copiedNow; - - /* - * The check below does too much because it will reject a call to this - * function with a channel which is part of an 'fcopy'. But we have to - * allow this here or else the chaining in the transformation drivers will - * fail with 'file busy' error instead of retrieving and transforming the - * data to copy. - * - * We let the check procedure now believe that there is no fcopy in - * progress. A better solution than this might be an additional flag - * argument to switch off specific checks. - */ + int copied = 0; if (CheckChannelErrors(statePtr, TCL_READABLE | CHANNEL_RAW_MODE) != 0) { return -1; } - /* - * Check for information in the push-back buffers. If there is some, use - * it. Go to the driver only if there is none (anymore) and the caller - * requests more bytes. - */ + /* First read bytes from the push-back buffers. */ - for (copied = 0; copied < bytesToRead; copied += copiedNow) { - copiedNow = CopyBuffer(chanPtr, bufPtr + copied, - bytesToRead - copied); - if (copiedNow == 0) { - if (GotFlag(statePtr, CHANNEL_EOF)) { - goto done; - } - if (GotFlag(statePtr, CHANNEL_BLOCKED)) { - if (GotFlag(statePtr, CHANNEL_NONBLOCKING)) { - goto done; - } - ResetFlag(statePtr, CHANNEL_BLOCKED); - } + while (chanPtr->inQueueHead && bytesToRead > 0) { + ChannelBuffer *bufPtr = chanPtr->inQueueHead; + int bytesInBuffer = BytesLeft(bufPtr); + int toCopy = (bytesInBuffer < bytesToRead) ? bytesInBuffer + : bytesToRead; -#ifdef TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING - /* - * [Bug 943274]. Better emulation of non-blocking channels for - * channels without BlockModeProc, by keeping track of true - * fileevents generated by the OS == Data waiting and reading if - * and only if we are sure to have data. - */ + /* Copy the current chunk into the read buffer. */ - if (GotFlag(statePtr, CHANNEL_NONBLOCKING) && - (Tcl_ChannelBlockModeProc(chanPtr->typePtr) == NULL) && - !GotFlag(statePtr, CHANNEL_HAS_MORE_DATA)) { - /* - * We bypass the driver; it would block as no data is - * available. - */ + memcpy(readBuf, RemovePoint(bufPtr), (size_t) toCopy); + bufPtr->nextRemoved += toCopy; + copied += toCopy; + readBuf += toCopy; + bytesToRead -= toCopy; - nread = -1; - result = EWOULDBLOCK; - } else -#endif /* TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING */ - { - /* - * Now go to the driver to get as much as is possible to fill - * the remaining request. Do all the error handling by - * ourselves. The code was stolen from 'GetInput' and slightly - * adapted (different return value here). - * - * The case of 'bytesToRead == 0' at this point cannot happen. - */ + /* If the current buffer is empty recycle it. */ - nread = ChanRead(chanPtr, bufPtr + copied, - bytesToRead - copied, &result); + if (IsBufferEmpty(bufPtr)) { + chanPtr->inQueueHead = bufPtr->nextPtr; + if (chanPtr->inQueueHead == NULL) { + chanPtr->inQueueTail = NULL; } + RecycleBuffer(chanPtr->state, bufPtr, 0); + } + } - if (nread > 0) { - /* - * If we get a short read, signal up that we may be BLOCKED. - * We should avoid calling the driver because on some - * platforms we will block in the low level reading code even - * though the channel is set into nonblocking mode. - */ - - if (nread < (bytesToRead - copied)) { - SetFlag(statePtr, CHANNEL_BLOCKED); - } - -#ifdef TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING - if (nread <= (bytesToRead - copied)) { - /* - * [Bug 943274] We have read the available data, clear - * flag. - */ - - ResetFlag(statePtr, CHANNEL_HAS_MORE_DATA); - } -#endif /* TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING */ - } else if (nread == 0) { - SetFlag(statePtr, CHANNEL_EOF); - statePtr->inputEncodingFlags |= TCL_ENCODING_END; - - } else if (nread < 0) { - if ((result == EWOULDBLOCK) || (result == EAGAIN)) { - if (copied > 0) { - /* - * Information that was copied earlier has precedence - * over EAGAIN/WOULDBLOCK handling. - */ + /* Go to the driver if more data needed. */ - return copied; - } + if (bytesToRead > 0) { - SetFlag(statePtr, CHANNEL_BLOCKED); - result = EAGAIN; - } + int nread = ChanRead(chanPtr, readBuf, bytesToRead); - Tcl_SetErrno(result); - return -1; + if (nread > 0) { + /* Successful read (short is OK) - add to bytes copied */ + copied += nread; + } else if (nread < 0) { + /* + * An error signaled. If CHANNEL_BLOCKED, then the error + * is not real, but an indication of blocked state. In + * that case, retain the flag and let caller receive the + * short read of copied bytes from the pushback. + * HOWEVER, if copied==0 bytes from pushback then repeat + * signalling the blocked state as an error to caller so + * there is no false report of an EOF. + * When !CHANNEL_BLOCKED, the error is real and passes on + * to caller. + */ + if (!GotFlag(statePtr, CHANNEL_BLOCKED) || copied == 0) { + copied = -1; } - - return copied + nread; + } else if (copied > 0) { + /* + * nread == 0. Driver is at EOF, but if copied>0 bytes + * from pushback, then we should not signal it yet. + */ + ResetFlag(statePtr, CHANNEL_EOF); } } - - done: return copied; } @@ -5738,8 +5569,9 @@ DoReadChars( ChannelState *statePtr = chanPtr->state; /* State info for channel */ ChannelBuffer *bufPtr; - int offset, factor, copied, copiedNow, result; + int factor, copied, copiedNow, result; Tcl_Encoding encoding; + int binaryMode; #define UTF_EXPANSION_FACTOR 1024 /* @@ -5749,9 +5581,14 @@ DoReadChars( chanPtr = statePtr->topChanPtr; encoding = statePtr->encoding; factor = UTF_EXPANSION_FACTOR; + Tcl_Preserve(chanPtr); + + binaryMode = (encoding == NULL) + && (statePtr->inputTranslation == TCL_TRANSLATE_LF) + && (statePtr->inEofChar == '\0'); if (appendFlag == 0) { - if (encoding == NULL) { + if (binaryMode) { Tcl_SetByteArrayLength(objPtr, 0); } else { Tcl_SetObjLength(objPtr, 0); @@ -5760,27 +5597,23 @@ DoReadChars( * We're going to access objPtr->bytes directly, so we must ensure * that this is actually a string object (otherwise it might have * been pure Unicode). + * + * Probably not needed anymore. */ TclGetString(objPtr); } - offset = 0; - } else { - if (encoding == NULL) { - Tcl_GetByteArrayFromObj(objPtr, &offset); - } else { - TclGetStringFromObj(objPtr, &offset); - } } + /* Must clear the BLOCKED flag here since we check before reading */ + ResetFlag(statePtr, CHANNEL_BLOCKED); for (copied = 0; (unsigned) toRead > 0; ) { copiedNow = -1; if (statePtr->inQueueHead != NULL) { - if (encoding == NULL) { - copiedNow = ReadBytes(statePtr, objPtr, toRead, &offset); + if (binaryMode) { + copiedNow = ReadBytes(statePtr, objPtr, toRead); } else { - copiedNow = ReadChars(statePtr, objPtr, toRead, &offset, - &factor); + copiedNow = ReadChars(statePtr, objPtr, toRead, &factor); } /* @@ -5803,19 +5636,21 @@ DoReadChars( if (GotFlag(statePtr, CHANNEL_EOF)) { break; } - if (GotFlag(statePtr, CHANNEL_BLOCKED)) { - if (GotFlag(statePtr, CHANNEL_NONBLOCKING)) { - break; - } - ResetFlag(statePtr, CHANNEL_BLOCKED); + if (GotFlag(statePtr, CHANNEL_NONBLOCKING|CHANNEL_BLOCKED) + == (CHANNEL_NONBLOCKING|CHANNEL_BLOCKED)) { + break; } result = GetInput(chanPtr); + if (chanPtr != statePtr->topChanPtr) { + Tcl_Release(chanPtr); + chanPtr = statePtr->topChanPtr; + Tcl_Preserve(chanPtr); + } if (result != 0) { - if (result == EAGAIN) { - break; + if (!GotFlag(statePtr, CHANNEL_BLOCKED)) { + copied = -1; } - copied = -1; - goto done; + break; } } else { copied += copiedNow; @@ -5823,27 +5658,31 @@ DoReadChars( } } - ResetFlag(statePtr, CHANNEL_BLOCKED); - if (encoding == NULL) { - Tcl_SetByteArrayLength(objPtr, offset); - } else { - Tcl_SetObjLength(objPtr, offset); - } - /* - * Update the notifier state so we don't block while there is still data - * in the buffers. + * Failure to fill a channel buffer may have left channel reporting + * a "blocked" state, but so long as we fulfilled the request here, + * the caller does not consider us blocked. */ + if (toRead == 0) { + ResetFlag(statePtr, CHANNEL_BLOCKED); + } - done: /* * Regenerate the top channel, in case it was changed due to * self-modifying reflected transforms. */ + if (chanPtr != statePtr->topChanPtr) { + Tcl_Release(chanPtr); + chanPtr = statePtr->topChanPtr; + Tcl_Preserve(chanPtr); + } - chanPtr = statePtr->topChanPtr; - + /* + * Update the notifier state so we don't block while there is still data + * in the buffers. + */ UpdateInterest(chanPtr); + Tcl_Release(chanPtr); return copied; } @@ -5861,13 +5700,11 @@ DoReadChars( * allocated to hold data read from the channel as needed. * * Results: - * The return value is the number of bytes appended to the object and - * *offsetPtr is filled with the total number of bytes in the object - * (greater than the return value if there were already bytes in the - * object). + * The return value is the number of bytes appended to the object, or + * -1 to indicate that zero bytes were read due to an EOF. * * Side effects: - * None. + * The storage of bytes in objPtr can cause (re-)allocation of memory. * *--------------------------------------------------------------------------- */ @@ -5880,72 +5717,22 @@ ReadBytes( * been allocated to hold data, not how many * bytes of data have been stored in the * object. */ - int bytesToRead, /* Maximum number of bytes to store, or < 0 to + int bytesToRead) /* Maximum number of bytes to store, or < 0 to * get all available bytes. Bytes are obtained * from the first buffer in the queue - even * if this number is larger than the number of * bytes available in the first buffer, only * the bytes from the first buffer are * returned. */ - int *offsetPtr) /* On input, contains how many bytes of objPtr - * have been used to hold data. On output, - * filled with how many bytes are now being - * used. */ { - int toRead, srcLen, offset, length, srcRead, dstWrote; - ChannelBuffer *bufPtr; - char *src, *dst; + ChannelBuffer *bufPtr = statePtr->inQueueHead; + int srcLen = BytesLeft(bufPtr); + int toRead = bytesToRead>srcLen || bytesToRead<0 ? srcLen : bytesToRead; - offset = *offsetPtr; - - bufPtr = statePtr->inQueueHead; - src = RemovePoint(bufPtr); - srcLen = BytesLeft(bufPtr); - - toRead = bytesToRead; - if ((unsigned) toRead > (unsigned) srcLen) { - toRead = srcLen; - } - - dst = (char *) Tcl_GetByteArrayFromObj(objPtr, &length); - if (toRead > length - offset - 1) { - /* - * Double the existing size of the object or make enough room to hold - * all the characters we may get from the source buffer, whichever is - * larger. - */ - - length = offset * 2; - if (offset < toRead) { - length = offset + toRead + 1; - } - dst = (char *) Tcl_SetByteArrayLength(objPtr, length); - } - dst += offset; - - if (GotFlag(statePtr, INPUT_NEED_NL)) { - ResetFlag(statePtr, INPUT_NEED_NL); - if ((srcLen == 0) || (*src != '\n')) { - *dst = '\r'; - *offsetPtr += 1; - return 1; - } - *dst++ = '\n'; - src++; - srcLen--; - toRead--; - } - - srcRead = srcLen; - dstWrote = toRead; - if (TranslateInputEOL(statePtr, dst, src, &dstWrote, &srcRead) != 0) { - if (dstWrote == 0) { - return -1; - } - } - bufPtr->nextRemoved += srcRead; - *offsetPtr += dstWrote; - return dstWrote; + TclAppendBytesToByteArray(objPtr, (unsigned char *) RemovePoint(bufPtr), + toRead); + bufPtr->nextRemoved += toRead; + return toRead; } /* @@ -5989,262 +5776,351 @@ ReadChars( * is larger than the number of characters * available in the first buffer, only the * characters from the first buffer are - * returned. */ - int *offsetPtr, /* On input, contains how many bytes of objPtr - * have been used to hold data. On output, - * filled with how many bytes are now being - * used. */ + * returned. The execption is when there is + * not any complete character in the first + * buffer. In that case, a recursive call + * effectively obtains chars from the + * second buffer. */ int *factorPtr) /* On input, contains a guess of how many * bytes need to be allocated to hold the * result of converting N source bytes to * UTF-8. On output, contains another guess * based on the data seen so far. */ { - int toRead, factor, offset, spaceLeft, srcLen, dstNeeded; - int srcRead, dstWrote, numChars, dstRead; - ChannelBuffer *bufPtr; - char *src, *dst; - Tcl_EncodingState oldState; - int encEndFlagSuppressed = 0; + Tcl_Encoding encoding = statePtr->encoding? statePtr->encoding + : GetBinaryEncoding(); + Tcl_EncodingState savedState = statePtr->inputEncodingState; + ChannelBuffer *bufPtr = statePtr->inQueueHead; + int savedIEFlags = statePtr->inputEncodingFlags; + int savedFlags = statePtr->flags; + char *dst, *src = RemovePoint(bufPtr); + int dstLimit, numBytes, srcLen = BytesLeft(bufPtr); - factor = *factorPtr; - offset = *offsetPtr; - - bufPtr = statePtr->inQueueHead; - src = RemovePoint(bufPtr); - srcLen = BytesLeft(bufPtr); + /* + * One src byte can yield at most one character. So when the + * number of src bytes we plan to read is less than the limit on + * character count to be read, clearly we will remain within that + * limit, and we can use the value of "srcLen" as a tighter limit + * for sizing receiving buffers. + */ - toRead = charsToRead; - if ((unsigned) toRead > (unsigned) srcLen) { - toRead = srcLen; - } + int toRead = ((charsToRead<0)||(charsToRead > srcLen)) ? srcLen : charsToRead; /* * 'factor' is how much we guess that the bytes in the source buffer will * expand when converted to UTF-8 chars. This guess comes from analyzing * how many characters were produced by the previous pass. */ + + int factor = *factorPtr; + int dstNeeded = TCL_UTF_MAX - 1 + toRead * factor / UTF_EXPANSION_FACTOR; - dstNeeded = TCL_UTF_MAX - 1 + toRead * factor / UTF_EXPANSION_FACTOR; - spaceLeft = objPtr->length - offset; - - if (dstNeeded > spaceLeft) { - /* - * Double the existing size of the object or make enough room to hold - * all the characters we want from the source buffer, whichever is - * larger. - */ - - int length = offset + ((offset < dstNeeded) ? dstNeeded : offset); - - if (Tcl_AttemptSetObjLength(objPtr, length) == 0) { - length = offset + dstNeeded; - if (Tcl_AttemptSetObjLength(objPtr, length) == 0) { - dstNeeded = TCL_UTF_MAX - 1 + toRead; - length = offset + dstNeeded; - Tcl_SetObjLength(objPtr, length); - } - } - spaceLeft = length - offset; - } + (void) TclGetStringFromObj(objPtr, &numBytes); + Tcl_AppendToObj(objPtr, NULL, dstNeeded); if (toRead == srcLen) { - /* - * Want to convert the whole buffer in one pass. If we have enough - * space, convert it using all available space in object rather than - * using the factor. - */ - - dstNeeded = spaceLeft; + unsigned int size; + dst = TclGetStringStorage(objPtr, &size) + numBytes; + dstNeeded = size - numBytes; + } else { + dst = TclGetString(objPtr) + numBytes; } - dst = objPtr->bytes + offset; /* - * [Bug 1462248]: The cause of the crash reported in this bug is this: - * - * - ReadChars, called with a single buffer, with a incomplete - * multi-byte character at the end (only the first byte of it). - * - Encoding translation fails, asks for more data - * - Data is read, and eof is reached, TCL_ENCODING_END (TEE) is set. - * - ReadChar is called again, converts the first buffer, but due to TEE - * it does not check for incomplete multi-byte data, and the character - * just after the end of the first buffer is a valid completion of the - * multi-byte header in the actual buffer. The conversion reads more - * characters from the buffer then present. This causes nextRemoved to - * overshoot nextAdded and the next reads compute a negative srcLen, - * cause further translations to fail, causing copying of data into the - * next buffer using bad arguments, causing the mecpy for to eventually - * fail. + * This routine is burdened with satisfying several constraints. + * It cannot append more than 'charsToRead` chars onto objPtr. + * This is measured after encoding and translation transformations + * are completed. There is no precise number of src bytes that can + * be associated with the limit. Yet, when we are done, we must know + * precisely the number of src bytes that were consumed to produce + * the appended chars, so that all subsequent bytes are left in + * the buffers for future read operations. * - * In the end it is a memory access bug spiraling out of control if the - * conditions are _just so_. And ultimate cause is that TEE is given to a - * conversion where it should not. TEE signals that this is the last - * buffer. Except in our case it is not. - * - * My solution is to suppress TEE if the first buffer is not the last. We - * will eventually need it given that EOF has been reached, but not right - * now. This is what the new flag "endEncSuppressFlag" is for. - * - * The bug in 'Tcl_Utf2UtfProc' where it read from memory behind the - * actual buffer has been fixed as well, and fixes the problem with the - * crash too, but this would still allow the generic layer to - * accidentially break a multi-byte sequence if the conditions are just - * right, because again the ExternalToUtf would be successful where it - * should not. + * The consequence is that we have no choice but to implement a + * "trial and error" approach, where in general we may need to + * perform transformations and copies multiple times to achieve + * a consistent set of results. This takes the shape of a loop. */ - if ((statePtr->inputEncodingFlags & TCL_ENCODING_END) && - (bufPtr->nextPtr != NULL)) { + dstLimit = dstNeeded + 1; + while (1) { + int dstDecoded, dstRead, dstWrote, srcRead, numChars; + /* - * TEE is set for a buffer which is not the last. Squash it for now, - * and restore it later, before yielding control to our caller. + * Perform the encoding transformation. Read no more than + * srcLen bytes, write no more than dstLimit bytes. */ - statePtr->inputEncodingFlags &= ~TCL_ENCODING_END; - encEndFlagSuppressed = 1; - } + int code = Tcl_ExternalToUtf(NULL, encoding, src, srcLen, + statePtr->inputEncodingFlags & (bufPtr->nextPtr + ? ~0 : ~TCL_ENCODING_END), &statePtr->inputEncodingState, + dst, dstLimit, &srcRead, &dstDecoded, &numChars); - oldState = statePtr->inputEncodingState; - if (GotFlag(statePtr, INPUT_NEED_NL)) { /* - * We want a '\n' because the last character we saw was '\r'. + * Perform the translation transformation in place. Read no more + * than the dstDecoded bytes the encoding transformation actually + * produced. Capture the number of bytes written in dstWrote. + * Capture the number of bytes actually consumed in dstRead. */ - ResetFlag(statePtr, INPUT_NEED_NL); - Tcl_ExternalToUtf(NULL, statePtr->encoding, src, srcLen, - statePtr->inputEncodingFlags, &statePtr->inputEncodingState, - dst, TCL_UTF_MAX + 1, &srcRead, &dstWrote, &numChars); - if ((dstWrote > 0) && (*dst == '\n')) { + dstWrote = dstLimit; + dstRead = dstDecoded; + TranslateInputEOL(statePtr, dst, dst, &dstWrote, &dstRead); + + if (dstRead < dstDecoded) { + /* - * The next char was a '\n'. Consume it and produce a '\n'. + * The encoding transformation produced bytes that the + * translation transformation did not consume. Why did + * this happen? */ - bufPtr->nextRemoved += srcRead; - } else { + if (statePtr->inEofChar && dst[dstRead] == statePtr->inEofChar) { + /* + * 1) There's an eof char set on the channel, and + * we saw it and stopped translating at that point. + * + * NOTE the bizarre spec of TranslateInputEOL in this case. + * Clearly the eof char had to be read in order to account + * for the stopping, but the value of dstRead does not + * include it. + * + * Also rather bizarre, our caller can only notice an + * EOF condition if we return the value -1 as the number + * of chars read. This forces us to perform a 2-call + * dance where the first call can read all the chars + * up to the eof char, and the second call is solely + * for consuming the encoded eof char then pointed at + * by src so that we can return that magic -1 value. + * This seems really wasteful, especially since + * the first decoding pass of each call is likely to + * decode many bytes beyond that eof char that's all we + * care about. + */ + + if (dstRead == 0) { + /* + * Curious choice in the eof char handling. We leave + * the eof char in the buffer. So, no need to compute + * a proper srcRead value. At this point, there + * are no chars before the eof char in the buffer. + */ + Tcl_SetObjLength(objPtr, numBytes); + return -1; + } + + { + /* + * There are chars leading the buffer before the eof + * char. Adjust the dstLimit so we go back and read + * only those and do not encounter the eof char this + * time. + */ + + dstLimit = dstRead + TCL_UTF_MAX; + statePtr->flags = savedFlags; + statePtr->inputEncodingFlags = savedIEFlags; + statePtr->inputEncodingState = savedState; + continue; + } + } + /* - * The next char was not a '\n'. Produce a '\r'. + * 2) The other way to read fewer bytes than are decoded + * is when the final byte is \r and we're in a CRLF + * translation mode so we cannot decide whether to + * record \r or \n yet. */ - *dst = '\r'; - } - statePtr->inputEncodingFlags &= ~TCL_ENCODING_START; - *offsetPtr += 1; + assert(dst[dstRead] == '\r'); + assert(statePtr->inputTranslation == TCL_TRANSLATE_CRLF); - if (encEndFlagSuppressed) { - statePtr->inputEncodingFlags |= TCL_ENCODING_END; - } - return 1; - } + if (dstWrote > 0) { + /* + * There are chars we can read before we hit the bare cr. + * Go back with a smaller dstLimit so we get them in the + * next pass, compute a matching srcRead, and don't end + * up back here in this call. + */ - Tcl_ExternalToUtf(NULL, statePtr->encoding, src, srcLen, - statePtr->inputEncodingFlags, &statePtr->inputEncodingState, dst, - dstNeeded + 1, &srcRead, &dstWrote, &numChars); + dstLimit = dstRead + TCL_UTF_MAX; + statePtr->flags = savedFlags; + statePtr->inputEncodingFlags = savedIEFlags; + statePtr->inputEncodingState = savedState; + continue; + } - if (encEndFlagSuppressed) { - statePtr->inputEncodingFlags |= TCL_ENCODING_END; - } + assert(dstWrote == 0); + assert(dstRead == 0); - if (srcRead == 0) { - /* - * Not enough bytes in src buffer to make a complete char. Copy the - * bytes to the next buffer to make a new contiguous string, then tell - * the caller to fill the buffer with more bytes. - */ + /* + * We decoded only the bare cr, and we cannot read a + * translated char from that alone. We have to know what's + * next. So why do we only have the one decoded char? + */ - ChannelBuffer *nextPtr; + if (code != TCL_OK) { + char buffer[TCL_UTF_MAX + 2]; + int read, decoded, count; + + /* + * Didn't get everything the buffer could offer + */ + + statePtr->flags = savedFlags; + statePtr->inputEncodingFlags = savedIEFlags; + statePtr->inputEncodingState = savedState; + + Tcl_ExternalToUtf(NULL, encoding, src, srcLen, + statePtr->inputEncodingFlags & (bufPtr->nextPtr + ? ~0 : ~TCL_ENCODING_END), &statePtr->inputEncodingState, + buffer, TCL_UTF_MAX + 2, &read, &decoded, &count); + + if (count == 2) { + if (buffer[1] == '\n') { + /* \r\n translate to \n */ + dst[0] = '\n'; + bufPtr->nextRemoved += read; + } else { + dst[0] = '\r'; + bufPtr->nextRemoved += srcRead; + } + + dst[1] = '\0'; + statePtr->inputEncodingFlags &= ~TCL_ENCODING_START; + + Tcl_SetObjLength(objPtr, numBytes + 1); + return 1; + } + + } else if (statePtr->flags & CHANNEL_EOF) { - nextPtr = bufPtr->nextPtr; - if (nextPtr == NULL) { - if (srcLen > 0) { /* - * There isn't enough data in the buffers to complete the next - * character, so we need to wait for more data before the next - * file event can be delivered. [Bug 478856] - * - * The exception to this is if the input buffer was completely - * empty before we tried to convert its contents. Nothing in, - * nothing out, and no incomplete character data. The - * conversion before the current one was complete. + * The bare \r is the only char and we will never read + * a subsequent char to make the determination. */ - SetFlag(statePtr, CHANNEL_NEED_MORE_DATA); + dst[0] = '\r'; + bufPtr->nextRemoved = bufPtr->nextAdded; + Tcl_SetObjLength(objPtr, numBytes + 1); + return 1; } - return -1; + + /* + * Revise the dstRead value so that the numChars calc + * below correctly computes zero characters read. + */ + + dstRead = numChars; + + /* FALL THROUGH - get more data (dstWrote == 0) */ } - /* - * Space is made at the beginning of the buffer to copy the previous - * unused bytes there. Check first if the buffer we are using actually - * has enough space at its beginning for the data we are copying. - * Because if not we will write over the buffer management - * information, especially the 'nextPtr'. - * - * Note that the BUFFER_PADDING (See AllocChannelBuffer) is used to - * prevent exactly this situation. I.e. it should never happen. - * Therefore it is ok to panic should it happen despite the - * precautions. + /* + * The translation transformation can only reduce the number + * of chars when it converts \r\n into \n. The reduction in + * the number of chars is the difference in bytes read and written. */ - if (nextPtr->nextRemoved - srcLen < 0) { - Tcl_Panic("Buffer Underflow, BUFFER_PADDING not enough"); - } + numChars -= (dstRead - dstWrote); - nextPtr->nextRemoved -= srcLen; - memcpy(RemovePoint(nextPtr), src, (size_t) srcLen); - RecycleBuffer(statePtr, bufPtr, 0); - statePtr->inQueueHead = nextPtr; - return ReadChars(statePtr, objPtr, charsToRead, offsetPtr, factorPtr); - } + if (charsToRead > 0 && numChars > charsToRead) { - dstRead = dstWrote; - if (TranslateInputEOL(statePtr, dst, dst, &dstWrote, &dstRead) != 0) { - /* - * Hit EOF char. How many bytes of src correspond to where the EOF was - * located in dst? Run the conversion again with an output buffer just - * big enough to hold the data so we can get the correct value for - * srcRead. - */ + /* + * We read more chars than allowed. Reset limits to + * prevent that and try again. Don't forget the extra + * padding of TCL_UTF_MAX - 1 bytes demanded by the + * Tcl_ExternalToUtf() call! + */ - if (dstWrote == 0) { - return -1; + dstLimit = Tcl_UtfAtIndex(dst, charsToRead + 1) + + TCL_UTF_MAX - 1 - dst; + statePtr->flags = savedFlags; + statePtr->inputEncodingFlags = savedIEFlags; + statePtr->inputEncodingState = savedState; + continue; } - statePtr->inputEncodingState = oldState; - Tcl_ExternalToUtf(NULL, statePtr->encoding, src, srcLen, - statePtr->inputEncodingFlags, &statePtr->inputEncodingState, - dst, dstRead + TCL_UTF_MAX, &srcRead, &dstWrote, &numChars); - TranslateInputEOL(statePtr, dst, dst, &dstWrote, &dstRead); - } - /* - * The number of characters that we got may be less than the number that - * we started with because "\r\n" sequences may have been turned into just - * '\n' in dst. - */ + if (dstWrote == 0) { + ChannelBuffer *nextPtr; - numChars -= dstRead - dstWrote; + /* We were not able to read any chars. */ - if ((unsigned) numChars > (unsigned) toRead) { - /* - * Got too many chars. - */ + assert (numChars == 0); - const char *eof = Tcl_UtfAtIndex(dst, toRead); + /* + * There is one situation where this is the correct final + * result. If the src buffer contains only a single \n + * byte, and we are in TCL_TRANSLATE_AUTO mode, and + * when the translation pass was made the INPUT_SAW_CR + * flag was set on the channel. In that case, the + * correct behavior is to consume that \n and produce the + * empty string. + */ - statePtr->inputEncodingState = oldState; - Tcl_ExternalToUtf(NULL, statePtr->encoding, src, srcLen, - statePtr->inputEncodingFlags, &statePtr->inputEncodingState, - dst, eof - dst + TCL_UTF_MAX, &srcRead, &dstWrote, &numChars); - dstRead = dstWrote; - TranslateInputEOL(statePtr, dst, dst, &dstWrote, &dstRead); - numChars -= (dstRead - dstWrote); - } - statePtr->inputEncodingFlags &= ~TCL_ENCODING_START; + if (dst[0] == '\n') { + assert(statePtr->inputTranslation == TCL_TRANSLATE_AUTO); + assert(dstRead == 1); + + goto consume; + } + + /* Otherwise, reading zero characters indicates there's + * something incomplete at the end of the src buffer. + * Maybe there were not enough src bytes to decode into + * a char. Maybe a lone \r could not be translated (crlf + * mode). Need to combine any unused src bytes we have + * in the first buffer with subsequent bytes to try again. + */ - bufPtr->nextRemoved += srcRead; - if (dstWrote > srcRead + 1) { - *factorPtr = dstWrote * UTF_EXPANSION_FACTOR / srcRead; + nextPtr = bufPtr->nextPtr; + + if (nextPtr == NULL) { + if (srcLen > 0) { + SetFlag(statePtr, CHANNEL_NEED_MORE_DATA); + } + Tcl_SetObjLength(objPtr, numBytes); + return -1; + } + + /* + * Space is made at the beginning of the buffer to copy the + * previous unused bytes there. Check first if the buffer we + * are using actually has enough space at its beginning for + * the data we are copying. Because if not we will write over + * the buffer management information, especially the 'nextPtr'. + * + * Note that the BUFFER_PADDING (See AllocChannelBuffer) is + * used to prevent exactly this situation. I.e. it should never + * happen. Therefore it is ok to panic should it happen despite + * the precautions. + */ + + if (nextPtr->nextRemoved - srcLen < 0) { + Tcl_Panic("Buffer Underflow, BUFFER_PADDING not enough"); + } + + nextPtr->nextRemoved -= srcLen; + memcpy(RemovePoint(nextPtr), src, (size_t) srcLen); + RecycleBuffer(statePtr, bufPtr, 0); + statePtr->inQueueHead = nextPtr; + Tcl_SetObjLength(objPtr, numBytes); + return ReadChars(statePtr, objPtr, charsToRead, factorPtr); + } + + statePtr->inputEncodingFlags &= ~TCL_ENCODING_START; + + consume: + bufPtr->nextRemoved += srcRead; + /* + * If this read contained multibyte characters, revise factorPtr + * so the next read will allocate bigger buffers. + */ + if (numChars && numChars < srcRead) { + *factorPtr = srcRead * UTF_EXPANSION_FACTOR / numChars; + } + Tcl_SetObjLength(objPtr, numBytes + dstWrote); + return numChars; } - *offsetPtr += dstWrote; - return numChars; } /* @@ -6265,7 +6141,7 @@ ReadChars( *--------------------------------------------------------------------------- */ -static int +static void TranslateInputEOL( ChannelState *statePtr, /* Channel being read, for EOL translation and * EOF character. */ @@ -6274,132 +6150,138 @@ TranslateInputEOL( * characters. */ const char *srcStart, /* Source characters. */ int *dstLenPtr, /* On entry, the maximum length of output - * buffer in bytes; must be <= *srcLenPtr. On - * exit, the number of bytes actually used in - * output buffer. */ + * buffer in bytes. On exit, the number of + * bytes actually used in output buffer. */ int *srcLenPtr) /* On entry, the length of source buffer. On * exit, the number of bytes read from the * source buffer. */ { - int dstLen, srcLen, inEofChar; - const char *eof; + const char *eof = NULL; + int dstLen = *dstLenPtr; + int srcLen = *srcLenPtr; + int inEofChar = statePtr->inEofChar; - dstLen = *dstLenPtr; + /* + * Depending on the translation mode in use, there's no need + * to scan more srcLen bytes at srcStart than can possibly transform + * to dstLen bytes. This keeps the scan for eof char below from + * being pointlessly long. + */ + + switch (statePtr->inputTranslation) { + case TCL_TRANSLATE_LF: + case TCL_TRANSLATE_CR: + if (srcLen > dstLen) { + /* In these modes, each src byte become a dst byte. */ + srcLen = dstLen; + } + break; + default: + /* In other modes, at most 2 src bytes become a dst byte. */ + if (srcLen > 2 * dstLen) { + srcLen = 2 * dstLen; + } + break; + } - eof = NULL; - inEofChar = statePtr->inEofChar; if (inEofChar != '\0') { /* - * Find EOF in translated buffer then compress out the EOL. The source - * buffer may be much longer than the destination buffer - we only - * want to return EOF if the EOF has been copied to the destination - * buffer. + * Make sure we do not read past any logical end of channel input + * created by the presence of the input eof char. */ - const char *src, *srcMax = srcStart + *srcLenPtr; - - for (src = srcStart; src < srcMax; src++) { - if (*src == inEofChar) { - eof = src; - srcLen = src - srcStart; - if (srcLen < dstLen) { - dstLen = srcLen; - } - *srcLenPtr = srcLen; - break; - } + if ((eof = memchr(srcStart, inEofChar, srcLen))) { + srcLen = eof - srcStart; } } + switch (statePtr->inputTranslation) { case TCL_TRANSLATE_LF: + case TCL_TRANSLATE_CR: if (dstStart != srcStart) { - memcpy(dstStart, srcStart, (size_t) dstLen); + memcpy(dstStart, srcStart, (size_t) srcLen); } - srcLen = dstLen; - break; - case TCL_TRANSLATE_CR: { - char *dst, *dstEnd; + if (statePtr->inputTranslation == TCL_TRANSLATE_CR) { + char *dst = dstStart; + char *dstEnd = dstStart + srcLen; - if (dstStart != srcStart) { - memcpy(dstStart, srcStart, (size_t) dstLen); - } - dstEnd = dstStart + dstLen; - for (dst = dstStart; dst < dstEnd; dst++) { - if (*dst == '\r') { - *dst = '\n'; + while ((dst = memchr(dst, '\r', dstEnd - dst))) { + *dst++ = '\n'; } } - srcLen = dstLen; + dstLen = srcLen; break; - } case TCL_TRANSLATE_CRLF: { - char *dst; - const char *src, *srcEnd, *srcMax; - - dst = dstStart; - src = srcStart; - srcEnd = srcStart + dstLen; - srcMax = srcStart + *srcLenPtr; - - for ( ; src < srcEnd; ) { - if (*src == '\r') { - src++; - if (src >= srcMax) { - SetFlag(statePtr, INPUT_NEED_NL); - } else if (*src == '\n') { - *dst++ = *src++; - } else { + const char *crFound, *src = srcStart; + char *dst = dstStart; + int lesser = (dstLen < srcLen) ? dstLen : srcLen; + + while ((crFound = memchr(src, '\r', lesser))) { + int numBytes = crFound - src; + memmove(dst, src, numBytes); + + dst += numBytes; dstLen -= numBytes; + src += numBytes; srcLen -= numBytes; + if (srcLen == 1) { + /* valid src bytes end in \r */ + if (eof) { *dst++ = '\r'; + src++; srcLen--; + } else { + lesser = 0; + break; } + } else if (src[1] == '\n') { + *dst++ = '\n'; + src += 2; srcLen -= 2; } else { - *dst++ = *src++; + *dst++ = '\r'; + src++; srcLen--; } + dstLen--; + lesser = (dstLen < srcLen) ? dstLen : srcLen; } - srcLen = src - srcStart; - dstLen = dst - dstStart; + memmove(dst, src, lesser); + srcLen = src + lesser - srcStart; + dstLen = dst + lesser - dstStart; break; } case TCL_TRANSLATE_AUTO: { - char *dst; - const char *src, *srcEnd, *srcMax; + const char *crFound, *src = srcStart; + char *dst = dstStart; + int lesser; - dst = dstStart; - src = srcStart; - srcEnd = srcStart + dstLen; - srcMax = srcStart + *srcLenPtr; - - if (GotFlag(statePtr, INPUT_SAW_CR) && (src < srcMax)) { - if (*src == '\n') { - src++; - } + if ((statePtr->flags & INPUT_SAW_CR) && srcLen) { + if (*src == '\n') { src++; srcLen--; } ResetFlag(statePtr, INPUT_SAW_CR); } - for ( ; src < srcEnd; ) { - if (*src == '\r') { - src++; - if (src >= srcMax) { - SetFlag(statePtr, INPUT_SAW_CR); - } else if (*src == '\n') { - if (srcEnd < srcMax) { - srcEnd++; - } - src++; - } - *dst++ = '\n'; - } else { - *dst++ = *src++; + lesser = (dstLen < srcLen) ? dstLen : srcLen; + while ((crFound = memchr(src, '\r', lesser))) { + int numBytes = crFound - src; + memmove(dst, src, numBytes); + + dst[numBytes] = '\n'; + dst += numBytes + 1; dstLen -= numBytes + 1; + src += numBytes + 1; srcLen -= numBytes + 1; + if (srcLen == 0) { + SetFlag(statePtr, INPUT_SAW_CR); + } else if (*src == '\n') { + src++; srcLen--; } + lesser = (dstLen < srcLen) ? dstLen : srcLen; } - srcLen = src - srcStart; - dstLen = dst - dstStart; + memmove(dst, src, lesser); + srcLen = src + lesser - srcStart; + dstLen = dst + lesser - dstStart; break; } default: - return 0; + Tcl_Panic("unknown input translation %d", statePtr->inputTranslation); } *dstLenPtr = dstLen; + *srcLenPtr = srcLen; - if ((eof != NULL) && (srcStart + srcLen >= eof)) { + if (srcStart + srcLen == eof) { /* * EOF character was seen in EOL translated range. Leave current file * position pointing at the EOF character, but don't store the EOF @@ -6408,12 +6290,8 @@ TranslateInputEOL( SetFlag(statePtr, CHANNEL_EOF | CHANNEL_STICKY_EOF); statePtr->inputEncodingFlags |= TCL_ENCODING_END; - ResetFlag(statePtr, INPUT_SAW_CR | INPUT_NEED_NL); - return 1; + ResetFlag(statePtr, INPUT_SAW_CR); } - - *srcLenPtr = srcLen; - return 0; } /* @@ -6467,16 +6345,11 @@ Tcl_Ungets( statePtr->flags = flags; /* - * If we have encountered a sticky EOF, just punt without storing (sticky - * EOF is set if we have seen the input eofChar, to prevent reading beyond - * the eofChar). Otherwise, clear the EOF flags, and clear the BLOCKED - * bit. We want to discover these conditions anew in each operation. + * Clear the EOF flags, and clear the BLOCKED bit. */ - if (GotFlag(statePtr, CHANNEL_STICKY_EOF)) { - goto done; - } - ResetFlag(statePtr, CHANNEL_BLOCKED | CHANNEL_EOF); + ResetFlag(statePtr, + CHANNEL_BLOCKED | CHANNEL_STICKY_EOF | CHANNEL_EOF | INPUT_SAW_CR); bufPtr = AllocChannelBuffer(len); memcpy(InsertPoint(bufPtr), str, (size_t) len); @@ -6541,14 +6414,6 @@ Tcl_Flush( return -1; } - /* - * Force current output buffer to be output also. - */ - - if ((statePtr->curOutPtr != NULL) && IsBufferReady(statePtr->curOutPtr)) { - SetFlag(statePtr, BUFFER_READY); - } - result = FlushChannel(NULL, chanPtr, 0); if (result != 0) { return TCL_ERROR; @@ -6599,7 +6464,7 @@ DiscardInputQueued( */ if (discardSavedBuffers && statePtr->saveInBufPtr != NULL) { - ckfree(statePtr->saveInBufPtr); + ReleaseChannelBuffer(statePtr->saveInBufPtr); statePtr->saveInBufPtr = NULL; } } @@ -6611,6 +6476,9 @@ DiscardInputQueued( * * Reads input data from a device into a channel buffer. * + * IMPORTANT! This routine is only called on a chanPtr argument + * that is the top channel of a stack! + * * Results: * The return value is the Posix error code if an error occurred while * reading from the file, or 0 otherwise. @@ -6643,21 +6511,28 @@ GetInput( return EINVAL; } + /* + * For a channel at EOF do not bother allocating buffers; there's + * nothing more to read. Avoid calling the driver inputproc in + * case some of them do not react well to additional calls after + * they've reported an eof state.. + * TODO: Candidate for a can't happen panic. + */ + + if (GotFlag(statePtr, CHANNEL_EOF)) { + return 0; + } + /* * First check for more buffers in the pushback area of the topmost * channel in the stack and use them. They can be the result of a * transformation which went away without reading all the information * placed in the area when it was stacked. - * - * Two possibilities for the state: No buffers in it, or a single empty - * buffer. In the latter case we can recycle it now. */ if (chanPtr->inQueueHead != NULL) { - if (statePtr->inQueueHead != NULL) { - RecycleBuffer(statePtr, statePtr->inQueueHead, 0); - statePtr->inQueueHead = NULL; - } + + assert(statePtr->inQueueHead == NULL); statePtr->inQueueHead = chanPtr->inQueueHead; statePtr->inQueueTail = chanPtr->inQueueTail; @@ -6678,21 +6553,20 @@ GetInput( */ bufPtr = statePtr->inQueueTail; - if ((bufPtr != NULL) && !IsBufferFull(bufPtr)) { - toRead = SpaceLeft(bufPtr); - } else { + + if ((bufPtr == NULL) || IsBufferFull(bufPtr)) { bufPtr = statePtr->saveInBufPtr; statePtr->saveInBufPtr = NULL; /* * Check the actual buffersize against the requested buffersize. - * Buffers which are smaller than requested are squashed. This is done + * Saved buffers of the wrong size are squashed. This is done * to honor dynamic changes of the buffersize made by the user. */ if ((bufPtr != NULL) - && (bufPtr->bufLength - BUFFER_PADDING < statePtr->bufSize)) { - ckfree(bufPtr); + && (bufPtr->bufLength - BUFFER_PADDING != statePtr->bufSize)) { + ReleaseChannelBuffer(bufPtr); bufPtr = NULL; } @@ -6701,21 +6575,8 @@ GetInput( } bufPtr->nextPtr = NULL; - /* - * SF #427196: Use the actual size of the buffer to determine the - * number of bytes to read from the channel and not the size for new - * buffers. They can be different if the buffersize was changed - * between reads. - * - * Note: This affects performance negatively if the buffersize was - * extended but this small buffer is reused for all subsequent reads. - * The system never uses buffers with the requested bigger size in - * that case. An adjunct patch could try and delete all unused buffers - * it encounters and which are smaller than the formally requested - * buffersize. - */ - toRead = SpaceLeft(bufPtr); + assert(toRead == statePtr->bufSize); if (statePtr->inQueueTail == NULL) { statePtr->inQueueHead = bufPtr; @@ -6723,75 +6584,22 @@ GetInput( statePtr->inQueueTail->nextPtr = bufPtr; } statePtr->inQueueTail = bufPtr; + } else { + toRead = SpaceLeft(bufPtr); } - /* - * If EOF is set, we should avoid calling the driver because on some - * platforms it is impossible to read from a device after EOF. - */ - - if (GotFlag(statePtr, CHANNEL_EOF)) { - return 0; - } - -#ifdef TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING - /* - * [Bug 943274]: Better emulation of non-blocking channels for channels - * without BlockModeProc, by keeping track of true fileevents generated by - * the OS == Data waiting and reading if and only if we are sure to have - * data. - */ - - if (GotFlag(statePtr, CHANNEL_NONBLOCKING) && - (Tcl_ChannelBlockModeProc(chanPtr->typePtr) == NULL) && - !GotFlag(statePtr, CHANNEL_HAS_MORE_DATA)) { - /* - * Bypass the driver, it would block, as no data is available - */ - - nread = -1; - result = EWOULDBLOCK; - } else -#endif /* TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING */ - { - nread = ChanRead(chanPtr, InsertPoint(bufPtr), toRead, &result); - } + PreserveChannelBuffer(bufPtr); + nread = ChanRead(chanPtr, InsertPoint(bufPtr), toRead); - if (nread > 0) { + if (nread < 0) { + result = Tcl_GetErrno(); + } else { + result = 0; bufPtr->nextAdded += nread; - - /* - * If we get a short read, signal up that we may be BLOCKED. We should - * avoid calling the driver because on some platforms we will block in - * the low level reading code even though the channel is set into - * nonblocking mode. - */ - - if (nread < toRead) { - SetFlag(statePtr, CHANNEL_BLOCKED); - } - -#ifdef TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING - if (nread <= toRead) { - /* - * [Bug 943274]: We have read the available data, clear flag. - */ - - ResetFlag(statePtr, CHANNEL_HAS_MORE_DATA); - } -#endif /* TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING */ - } else if (nread == 0) { - SetFlag(statePtr, CHANNEL_EOF); - statePtr->inputEncodingFlags |= TCL_ENCODING_END; - } else if (nread < 0) { - if ((result == EWOULDBLOCK) || (result == EAGAIN)) { - SetFlag(statePtr, CHANNEL_BLOCKED); - result = EAGAIN; - } - Tcl_SetErrno(result); - return result; } - return 0; + + ReleaseChannelBuffer(bufPtr); + return result; } /* @@ -6919,15 +6727,6 @@ Tcl_Seek( } /* - * If there is data buffered in statePtr->curOutPtr then mark the channel - * as ready to flush before invoking FlushChannel. - */ - - if ((statePtr->curOutPtr != NULL) && IsBufferReady(statePtr->curOutPtr)) { - SetFlag(statePtr, BUFFER_READY); - } - - /* * If the flush fails we cannot recover the original position. In that * case the seek is not attempted because we do not know where the access * position is - instead we return the error. FlushChannel has already @@ -7251,17 +7050,7 @@ CheckChannelErrors( } if (direction == TCL_READABLE) { - /* - * If we have not encountered a sticky EOF, clear the EOF bit (sticky - * EOF is set if we have seen the input eofChar, to prevent reading - * beyond the eofChar). Also, always clear the BLOCKED bit. We want to - * discover these conditions anew in each operation. - */ - - if (!GotFlag(statePtr, CHANNEL_STICKY_EOF)) { - ResetFlag(statePtr, CHANNEL_EOF); - } - ResetFlag(statePtr, CHANNEL_BLOCKED | CHANNEL_NEED_MORE_DATA); + ResetFlag(statePtr, CHANNEL_NEED_MORE_DATA); } return 0; @@ -7477,14 +7266,26 @@ Tcl_SetChannelBufferSize( } statePtr = ((Channel *) chan)->state; + + if (statePtr->bufSize == sz) { + return; + } statePtr->bufSize = sz; - if (statePtr->outputStage != NULL) { - ckfree(statePtr->outputStage); - statePtr->outputStage = NULL; + /* + * If bufsize changes, need to get rid of old utility buffer. + */ + + if (statePtr->saveInBufPtr != NULL) { + RecycleBuffer(statePtr, statePtr->saveInBufPtr, 1); + statePtr->saveInBufPtr = NULL; } - if ((statePtr->encoding != NULL) && GotFlag(statePtr, TCL_WRITABLE)) { - statePtr->outputStage = ckalloc(statePtr->bufSize + 2); + if ((statePtr->inQueueHead != NULL) + && (statePtr->inQueueHead->nextPtr == NULL) + && IsBufferEmpty(statePtr->inQueueHead)) { + RecycleBuffer(statePtr, statePtr->inQueueHead, 1); + statePtr->inQueueHead = NULL; + statePtr->inQueueTail = NULL; } } @@ -7920,6 +7721,7 @@ Tcl_SetChannelOption( return TCL_ERROR; } Tcl_SetChannelBufferSize(chan, newBufferSize); + return TCL_OK; } else if (HaveOpt(2, "-encoding")) { Tcl_Encoding encoding; @@ -7937,7 +7739,8 @@ Tcl_SetChannelOption( * iso2022, the terminated escape sequence must write to the buffer. */ - if ((statePtr->encoding != NULL) && (statePtr->curOutPtr != NULL) + if ((statePtr->encoding != NULL) + && !(statePtr->outputEncodingFlags & TCL_ENCODING_START) && (CheckChannelErrors(statePtr, TCL_WRITABLE) == 0)) { statePtr->outputEncodingFlags |= TCL_ENCODING_END; WriteChars(chanPtr, "", 0); @@ -7950,6 +7753,7 @@ Tcl_SetChannelOption( statePtr->outputEncodingFlags = TCL_ENCODING_START; ResetFlag(statePtr, CHANNEL_NEED_MORE_DATA); UpdateInterest(chanPtr); + return TCL_OK; } else if (HaveOpt(2, "-eofchar")) { if (Tcl_SplitList(interp, newValue, &argc, &argv) == TCL_ERROR) { return TCL_ERROR; @@ -8110,33 +7914,6 @@ Tcl_SetChannelOption( return Tcl_BadChannelOption(interp, optionName, NULL); } - /* - * If bufsize changes, need to get rid of old utility buffer. - */ - - if (statePtr->saveInBufPtr != NULL) { - RecycleBuffer(statePtr, statePtr->saveInBufPtr, 1); - statePtr->saveInBufPtr = NULL; - } - if ((statePtr->inQueueHead != NULL) - && (statePtr->inQueueHead->nextPtr == NULL) - && IsBufferEmpty(statePtr->inQueueHead)) { - RecycleBuffer(statePtr, statePtr->inQueueHead, 1); - statePtr->inQueueHead = NULL; - statePtr->inQueueTail = NULL; - } - - /* - * If encoding or bufsize changes, need to update output staging buffer. - */ - - if (statePtr->outputStage != NULL) { - ckfree(statePtr->outputStage); - statePtr->outputStage = NULL; - } - if ((statePtr->encoding != NULL) && GotFlag(statePtr, TCL_WRITABLE)) { - statePtr->outputStage = ckalloc(statePtr->bufSize + 2); - } return TCL_OK; } @@ -8228,21 +8005,6 @@ Tcl_NotifyChannel( Channel *upChanPtr; const Tcl_ChannelType *upTypePtr; -#ifdef TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING - /* - * [SF Tcl Bug 943274] For a non-blocking channel without blockmodeproc we - * keep track of actual input coming from the OS so that we can do a - * credible imitation of non-blocking behaviour. - */ - - if ((mask & TCL_READABLE) && - GotFlag(statePtr, CHANNEL_NONBLOCKING) && - (Tcl_ChannelBlockModeProc(chanPtr->typePtr) == NULL) && - !GotFlag(statePtr, CHANNEL_TIMER_FEV)) { - SetFlag(statePtr, CHANNEL_HAS_MORE_DATA); - } -#endif /* TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING */ - /* * In contrast to the other API functions this procedure walks towards the * top of a stack and not down from it. @@ -8302,8 +8064,9 @@ Tcl_NotifyChannel( */ if (GotFlag(statePtr, BG_FLUSH_SCHEDULED) && (mask & TCL_WRITABLE)) { - FlushChannel(NULL, chanPtr, 1); - mask &= ~TCL_WRITABLE; + if (0 == FlushChannel(NULL, chanPtr, 1)) { + mask &= ~TCL_WRITABLE; + } } /* @@ -8371,6 +8134,11 @@ UpdateInterest( /* State info for channel */ int mask = statePtr->interestMask; + if (chanPtr->typePtr == NULL) { + /* Do not update interest on a closed channel */ + return; + } + /* * If there are flushed buffers waiting to be written, then we need to * watch for the channel to become writable. @@ -8478,29 +8246,8 @@ ChannelTimerProc( statePtr->timer = Tcl_CreateTimerHandler(SYNTHETIC_EVENT_TIME, ChannelTimerProc,chanPtr); - -#ifdef TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING - /* - * Set the TIMER flag to notify the higher levels that the driver - * might have no data for us. We do this only if we are in - * non-blocking mode and the driver has no BlockModeProc because only - * then we really don't know if the driver will block or not. A - * similar test is done in "PeekAhead". - */ - - if (GotFlag(statePtr, CHANNEL_NONBLOCKING) && - (Tcl_ChannelBlockModeProc(chanPtr->typePtr) == NULL)) { - SetFlag(statePtr, CHANNEL_TIMER_FEV); - } -#endif /* TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING */ - Tcl_Preserve(statePtr); Tcl_NotifyChannel((Tcl_Channel) chanPtr, TCL_READABLE); - -#ifdef TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING - ResetFlag(statePtr, CHANNEL_TIMER_FEV); -#endif /* TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING */ - Tcl_Release(statePtr); } else { statePtr->timer = NULL; @@ -9283,9 +9030,9 @@ CopyData( } if (outBinary || sameEncoding) { - sizeb = DoWrite(outStatePtr->topChanPtr, buffer, sizeb); + sizeb = WriteBytes(outStatePtr->topChanPtr, buffer, sizeb); } else { - sizeb = DoWriteChars(outStatePtr->topChanPtr, buffer, sizeb); + sizeb = WriteChars(outStatePtr->topChanPtr, buffer, sizeb); } /* @@ -9440,12 +9187,23 @@ CopyData( * * DoRead -- * - * Reads a given number of bytes from a channel. No encoding conversions + * Stores up to "bytesToRead" bytes in memory pointed to by "dst". + * These bytes come from reading the channel "chanPtr" and + * performing the configured translations. No encoding conversions * are applied to the bytes being read. * * Results: - * The number of characters read, or -1 on error. Use Tcl_GetErrno() to - * retrieve the error code for the error that occurred. + * The number of bytes actually stored (<= bytesToRead), + * or -1 if there is an error in reading the channel. Use + * Tcl_GetErrno() to retrieve the error code for the error + * that occurred. + * + * The number of bytes stored can be less than the number + * requested when + * - EOF is reached on the channel; or + * - the channel is non-blocking, and we've read all we can + * without blocking. + * - a channel reading error occurs (and we return -1) * * Side effects: * May cause input to be buffered. @@ -9456,538 +9214,152 @@ CopyData( static int DoRead( Channel *chanPtr, /* The channel from which to read. */ - char *bufPtr, /* Where to store input read. */ - int toRead, /* Maximum number of bytes to read. */ + char *dst, /* Where to store input read. */ + int bytesToRead, /* Maximum number of bytes to read. */ int allowShortReads) /* Allow half-blocking (pipes,sockets) */ { ChannelState *statePtr = chanPtr->state; - /* State info for channel */ - int copied; /* How many characters were copied into the - * result string? */ - int copiedNow; /* How many characters were copied from the - * current input buffer? */ - int result; /* Of calling GetInput. */ - - /* - * If we have not encountered a sticky EOF, clear the EOF bit. Either way - * clear the BLOCKED bit. We want to discover these anew during each - * operation. - */ - - if (!GotFlag(statePtr, CHANNEL_STICKY_EOF)) { - ResetFlag(statePtr, CHANNEL_EOF); - } - ResetFlag(statePtr, CHANNEL_BLOCKED | CHANNEL_NEED_MORE_DATA); - - for (copied = 0; copied < toRead; copied += copiedNow) { - copiedNow = CopyAndTranslateBuffer(statePtr, bufPtr + copied, - toRead - copied); - if (copiedNow == 0) { - if (GotFlag(statePtr, CHANNEL_EOF)) { - goto done; - } - if (GotFlag(statePtr, CHANNEL_BLOCKED)) { - if (GotFlag(statePtr, CHANNEL_NONBLOCKING)) { - goto done; - } - ResetFlag(statePtr, CHANNEL_BLOCKED); - } - result = GetInput(chanPtr); - if (result != 0) { - if (result != EAGAIN) { - copied = -1; - } - goto done; - } - } else if (allowShortReads) { - copied += copiedNow; - break; - } - } - - ResetFlag(statePtr, CHANNEL_BLOCKED); - - /* - * Update the notifier state so we don't block while there is still data - * in the buffers. - */ - - done: - UpdateInterest(chanPtr); - return copied; -} - -/* - *---------------------------------------------------------------------- - * - * CopyAndTranslateBuffer -- - * - * Copy at most one buffer of input to the result space, doing eol - * translations according to mode in effect currently. - * - * Results: - * Number of bytes stored in the result buffer (as opposed to the number - * of bytes read from the channel). May return zero if no input is - * available to be translated. - * - * Side effects: - * Consumes buffered input. May deallocate one buffer. - * - *---------------------------------------------------------------------- - */ - -static int -CopyAndTranslateBuffer( - ChannelState *statePtr, /* Channel state from which to read input. */ - char *result, /* Where to store the copied input. */ - int space) /* How many bytes are available in result to - * store the copied input? */ -{ - ChannelBuffer *bufPtr; /* The buffer from which to copy bytes. */ - int bytesInBuffer; /* How many bytes are available to be copied - * in the current input buffer? */ - int copied; /* How many characters were already copied - * into the destination space? */ - int i; /* Iterates over the copied input looking for - * the input eofChar. */ - - /* - * If there is no input at all, return zero. The invariant is that either - * there is no buffer in the queue, or if the first buffer is empty, it is - * also the last buffer (and thus there is no input in the queue). Note - * also that if the buffer is empty, we leave it in the queue. - */ - - if (statePtr->inQueueHead == NULL) { - return 0; - } - bufPtr = statePtr->inQueueHead; - bytesInBuffer = BytesLeft(bufPtr); - - copied = 0; - switch (statePtr->inputTranslation) { - case TCL_TRANSLATE_LF: - if (bytesInBuffer == 0) { - return 0; - } + char *p = dst; + Tcl_Preserve(chanPtr); + while (bytesToRead) { /* - * Copy the current chunk into the result buffer. + * Each pass through the loop is intended to process up to + * one channel buffer. */ - if (bytesInBuffer < space) { - space = bytesInBuffer; - } - memcpy(result, RemovePoint(bufPtr), (size_t) space); - bufPtr->nextRemoved += space; - copied = space; - break; - case TCL_TRANSLATE_CR: { - char *end; - - if (bytesInBuffer == 0) { - return 0; - } + int bytesRead, bytesWritten; + ChannelBuffer *bufPtr = statePtr->inQueueHead; /* - * Copy the current chunk into the result buffer, then replace all \r - * with \n. + * When there's no buffered data to read, and we're at EOF, + * escape to the caller. */ - if (bytesInBuffer < space) { - space = bytesInBuffer; + if (statePtr->flags & CHANNEL_EOF + && (bufPtr == NULL || IsBufferEmpty(bufPtr))) { + break; } - memcpy(result, RemovePoint(bufPtr), (size_t) space); - bufPtr->nextRemoved += space; - copied = space; - for (end = result + copied; result < end; result++) { - if (*result == '\r') { - *result = '\n'; - } - } - break; - } - case TCL_TRANSLATE_CRLF: { - char *src, *end, *dst; - int curByte; + /* If there is no full buffer, attempt to create and/or fill one. */ - /* - * If there is a held-back "\r" at EOF, produce it now. - */ + while (!IsBufferFull(bufPtr)) { + int code; - if (bytesInBuffer == 0) { - if ((statePtr->flags & (INPUT_SAW_CR | CHANNEL_EOF)) == - (INPUT_SAW_CR | CHANNEL_EOF)) { - result[0] = '\r'; - ResetFlag(statePtr, INPUT_SAW_CR); - return 1; - } - return 0; - } - - /* - * Copy the current chunk and replace "\r\n" with "\n" (but not - * standalone "\r"!). - */ + moreData: + code = GetInput(chanPtr); + bufPtr = statePtr->inQueueHead; - if (bytesInBuffer < space) { - space = bytesInBuffer; - } - memcpy(result, RemovePoint(bufPtr), (size_t) space); - bufPtr->nextRemoved += space; - copied = space; + assert (bufPtr != NULL); - end = result + copied; - dst = result; - for (src = result; src < end; src++) { - curByte = *src; - if (curByte == '\n') { - ResetFlag(statePtr, INPUT_SAW_CR); - } else if (GotFlag(statePtr, INPUT_SAW_CR)) { - ResetFlag(statePtr, INPUT_SAW_CR); - *dst = '\r'; - dst++; + if (statePtr->flags & (CHANNEL_EOF|CHANNEL_BLOCKED)) { + /* Further reads cannot do any more */ + break; } - if (curByte == '\r') { - SetFlag(statePtr, INPUT_SAW_CR); - } else { - *dst = (char) curByte; - dst++; + + if (code) { + /* Read error */ + UpdateInterest(chanPtr); + Tcl_Release(chanPtr); + return -1; } - } - copied = dst - result; - break; - } - case TCL_TRANSLATE_AUTO: { - char *src, *end, *dst; - int curByte; - if (bytesInBuffer == 0) { - return 0; + assert (IsBufferFull(bufPtr)); } - /* - * Loop over the current buffer, converting "\r" and "\r\n" to "\n". - */ + assert (bufPtr != NULL); - if (bytesInBuffer < space) { - space = bytesInBuffer; - } - memcpy(result, RemovePoint(bufPtr), (size_t) space); - bufPtr->nextRemoved += space; - copied = space; + bytesRead = BytesLeft(bufPtr); + bytesWritten = bytesToRead; - end = result + copied; - dst = result; - for (src = result; src < end; src++) { - curByte = *src; - if (curByte == '\r') { - SetFlag(statePtr, INPUT_SAW_CR); - *dst = '\n'; - dst++; - } else { - if ((curByte != '\n') || !GotFlag(statePtr, INPUT_SAW_CR)) { - *dst = (char) curByte; - dst++; - } - ResetFlag(statePtr, INPUT_SAW_CR); - } - } - copied = dst - result; - break; - } - default: - Tcl_Panic("unknown eol translation mode"); - } + TranslateInputEOL(statePtr, p, RemovePoint(bufPtr), + &bytesWritten, &bytesRead); + bufPtr->nextRemoved += bytesRead; + p += bytesWritten; + bytesToRead -= bytesWritten; - /* - * If an in-stream EOF character is set for this channel, check that the - * input we copied so far does not contain the EOF char. If it does, copy - * only up to and excluding that character. - */ - - if (statePtr->inEofChar != 0) { - for (i = 0; i < copied; i++) { - if (result[i] == (char) statePtr->inEofChar) { - /* - * Set sticky EOF so that no further input is presented to the - * caller. - */ + if (!IsBufferEmpty(bufPtr)) { + /* + * Buffer is not empty. How can that be? + * + * 0) We stopped early because we got all the bytes + * we were seeking. That's fine. + */ - SetFlag(statePtr, CHANNEL_EOF | CHANNEL_STICKY_EOF); - statePtr->inputEncodingFlags |= TCL_ENCODING_END; - copied = i; + if (bytesToRead == 0) { + UpdateInterest(chanPtr); break; } - } - } - - /* - * If the current buffer is empty recycle it. - */ - - if (IsBufferEmpty(bufPtr)) { - statePtr->inQueueHead = bufPtr->nextPtr; - if (statePtr->inQueueHead == NULL) { - statePtr->inQueueTail = NULL; - } - RecycleBuffer(statePtr, bufPtr, 0); - } - - /* - * Return the number of characters copied into the result buffer. This may - * be different from the number of bytes consumed, because of EOL - * translations. - */ - return copied; -} - -/* - *---------------------------------------------------------------------- - * - * CopyBuffer -- - * - * Copy at most one buffer of input to the result space. - * - * Results: - * Number of bytes stored in the result buffer. May return zero if no - * input is available. - * - * Side effects: - * Consumes buffered input. May deallocate one buffer. - * - *---------------------------------------------------------------------- - */ - -static int -CopyBuffer( - Channel *chanPtr, /* Channel from which to read input. */ - char *result, /* Where to store the copied input. */ - int space) /* How many bytes are available in result to - * store the copied input? */ -{ - ChannelBuffer *bufPtr; /* The buffer from which to copy bytes. */ - int bytesInBuffer; /* How many bytes are available to be copied - * in the current input buffer? */ - int copied; /* How many characters were already copied - * into the destination space? */ - - /* - * If there is no input at all, return zero. The invariant is that either - * there is no buffer in the queue, or if the first buffer is empty, it is - * also the last buffer (and thus there is no input in the queue). Note - * also that if the buffer is empty, we don't leave it in the queue, but - * recycle it. - */ - - if (chanPtr->inQueueHead == NULL) { - return 0; - } - bufPtr = chanPtr->inQueueHead; - bytesInBuffer = BytesLeft(bufPtr); - - copied = 0; - - if (bytesInBuffer == 0) { - RecycleBuffer(chanPtr->state, bufPtr, 0); - chanPtr->inQueueHead = NULL; - chanPtr->inQueueTail = NULL; - return 0; - } - - /* - * Copy the current chunk into the result buffer. - */ - - if (bytesInBuffer < space) { - space = bytesInBuffer; - } - - memcpy(result, RemovePoint(bufPtr), (size_t) space); - bufPtr->nextRemoved += space; - copied = space; - - /* - * We don't care about in-stream EOF characters here as the data read here - * may still flow through one or more transformations, i.e. is not in its - * final state yet. - */ - - /* - * If the current buffer is empty recycle it. - */ - - if (IsBufferEmpty(bufPtr)) { - chanPtr->inQueueHead = bufPtr->nextPtr; - if (chanPtr->inQueueHead == NULL) { - chanPtr->inQueueTail = NULL; - } - RecycleBuffer(chanPtr->state, bufPtr, 0); - } - - /* - * Return the number of characters copied into the result buffer. - */ - - return copied; -} - -/* - *---------------------------------------------------------------------- - * - * DoWrite -- - * - * Puts a sequence of characters into an output buffer, may queue the - * buffer for output if it gets full, and also remembers whether the - * current buffer is ready e.g. if it contains a newline and we are in - * line buffering mode. - * - * Results: - * The number of bytes written or -1 in case of error. If -1, - * Tcl_GetErrno will return the error code. - * - * Side effects: - * May buffer up output and may cause output to be produced on the - * channel. - * - *---------------------------------------------------------------------- - */ - -static int -DoWrite( - Channel *chanPtr, /* The channel to buffer output for. */ - const char *src, /* Data to write. */ - int srcLen) /* Number of bytes to write. */ -{ - ChannelState *statePtr = chanPtr->state; - /* State info for channel */ - ChannelBuffer *outBufPtr; /* Current output buffer. */ - int foundNewline; /* Did we find a newline in output? */ - char *dPtr; - const char *sPtr; /* Search variables for newline. */ - int crsent; /* In CRLF eol translation mode, remember the - * fact that a CR was output to the channel - * without its following NL. */ - int i; /* Loop index for newline search. */ - int destCopied; /* How many bytes were used in this - * destination buffer to hold the output? */ - int totalDestCopied; /* How many bytes total were copied to the - * channel buffer? */ - int srcCopied; /* How many bytes were copied from the source - * string? */ - char *destPtr; /* Where in line to copy to? */ - - /* - * If we are in network (or windows) translation mode, record the fact - * that we have not yet sent a CR to the channel. - */ - - crsent = 0; - - /* - * Loop filling buffers and flushing them until all output has been - * consumed. - */ + /* + * 1) We're @EOF because we saw eof char. + */ - srcCopied = 0; - totalDestCopied = 0; + if (statePtr->inEofChar + && RemovePoint(bufPtr)[0] == statePtr->inEofChar) { + UpdateInterest(chanPtr); + break; + } - while (srcLen > 0) { - /* - * Make sure there is a current output buffer to accept output. - */ + /* + * 2) The buffer holds a \r while in CRLF translation, + * followed by the end of the buffer. + */ - if (statePtr->curOutPtr == NULL) { - statePtr->curOutPtr = AllocChannelBuffer(statePtr->bufSize); - } + assert(statePtr->inputTranslation == TCL_TRANSLATE_CRLF); + assert(RemovePoint(bufPtr)[0] == '\r'); + assert(BytesLeft(bufPtr) == 1); - outBufPtr = statePtr->curOutPtr; + if (bufPtr->nextPtr == NULL) { + /* There's no more buffered data.... */ - destCopied = SpaceLeft(outBufPtr); - if (destCopied > srcLen) { - destCopied = srcLen; - } + if (statePtr->flags & CHANNEL_EOF) { + /* ...and there never will be. */ - destPtr = InsertPoint(outBufPtr); - switch (statePtr->outputTranslation) { - case TCL_TRANSLATE_LF: - srcCopied = destCopied; - memcpy(destPtr, src, (size_t) destCopied); - break; - case TCL_TRANSLATE_CR: - srcCopied = destCopied; - memcpy(destPtr, src, (size_t) destCopied); - for (dPtr = destPtr; dPtr < destPtr + destCopied; dPtr++) { - if (*dPtr == '\n') { - *dPtr = '\r'; - } - } - break; - case TCL_TRANSLATE_CRLF: - for (srcCopied = 0, dPtr = destPtr, sPtr = src; - dPtr < destPtr + destCopied; - dPtr++, sPtr++, srcCopied++) { - if (*sPtr == '\n') { - if (crsent) { - *dPtr = '\n'; - crsent = 0; - } else { - *dPtr = '\r'; - crsent = 1; - sPtr--, srcCopied--; - } + *p++ = '\r'; + bytesToRead--; + bufPtr->nextRemoved++; + } else if (statePtr->flags & CHANNEL_BLOCKED) { + /* ...and we cannot get more now. */ + SetFlag(statePtr, CHANNEL_NEED_MORE_DATA); + UpdateInterest(chanPtr); + break; } else { - *dPtr = *sPtr; + /* ... so we need to get some. */ + goto moreData; } } - break; - case TCL_TRANSLATE_AUTO: - Tcl_Panic("Tcl_Write: AUTO output translation mode not supported"); - default: - Tcl_Panic("Tcl_Write: unknown output translation mode"); - } - /* - * The current buffer is ready for output if it is full, or if it - * contains a newline and this channel is line-buffered, or if it - * contains any output and this channel is unbuffered. - */ + if (bufPtr->nextPtr) { + /* There's a next buffer. Shift orphan \r to it. */ - outBufPtr->nextAdded += destCopied; - if (!GotFlag(statePtr, BUFFER_READY)) { - if (IsBufferFull(outBufPtr)) { - SetFlag(statePtr, BUFFER_READY); - } else if (GotFlag(statePtr, CHANNEL_LINEBUFFERED)) { - for (sPtr = src, i = 0, foundNewline = 0; - (i < srcCopied) && (!foundNewline); - i++, sPtr++) { - if (*sPtr == '\n') { - foundNewline = 1; - break; - } - } - if (foundNewline) { - SetFlag(statePtr, BUFFER_READY); - } - } else if (GotFlag(statePtr, CHANNEL_UNBUFFERED)) { - SetFlag(statePtr, BUFFER_READY); + ChannelBuffer *nextPtr = bufPtr->nextPtr; + + nextPtr->nextRemoved -= 1; + RemovePoint(nextPtr)[0] = '\r'; + bufPtr->nextRemoved++; } } - totalDestCopied += srcCopied; - src += srcCopied; - srcLen -= srcCopied; - - if (GotFlag(statePtr, BUFFER_READY)) { - if (FlushChannel(NULL, chanPtr, 0) != 0) { - return -1; + if (IsBufferEmpty(bufPtr)) { + statePtr->inQueueHead = bufPtr->nextPtr; + if (statePtr->inQueueHead == NULL) { + statePtr->inQueueTail = NULL; } + RecycleBuffer(statePtr, bufPtr, 0); } - } /* Closes "while" */ - return totalDestCopied; + if ((GotFlag(statePtr, CHANNEL_NONBLOCKING) || allowShortReads) + && GotFlag(statePtr, CHANNEL_BLOCKED)) { + break; + } + } + if (bytesToRead == 0) { + ResetFlag(statePtr, CHANNEL_BLOCKED); + } + + Tcl_Release(chanPtr); + return (int)(p - dst); } /* @@ -10107,12 +9479,15 @@ StackSetBlockMode( { int result = 0; Tcl_DriverBlockModeProc *blockModeProc; + ChannelState *statePtr = chanPtr->state; /* * Start at the top of the channel stack + * TODO: Examine what can go wrong when blockModeProc calls + * disturb the stacking state of the channel. */ - chanPtr = chanPtr->state->topChanPtr; + chanPtr = statePtr->topChanPtr; while (chanPtr != NULL) { blockModeProc = Tcl_ChannelBlockModeProc(chanPtr->typePtr); if (blockModeProc != NULL) { @@ -11215,12 +10590,11 @@ DupChannelIntRep( * currently have an internal rep.*/ { ChannelState *statePtr = GET_CHANNELSTATE(srcPtr); - Interp *interpPtr = GET_CHANNELINTERP(srcPtr); SET_CHANNELSTATE(copyPtr, statePtr); - SET_CHANNELINTERP(copyPtr, interpPtr); + SET_CHANNELINTERP(copyPtr, GET_CHANNELINTERP(srcPtr)); Tcl_Preserve(statePtr); - copyPtr->typePtr = &tclChannelType; + copyPtr->typePtr = srcPtr->typePtr; } /* @@ -11246,43 +10620,30 @@ SetChannelFromAny( register Tcl_Obj *objPtr) /* The object to convert. */ { ChannelState *statePtr; - Interp *interpPtr; if (interp == NULL) { return TCL_ERROR; } - if (objPtr->typePtr == &tclChannelType) { + if (objPtr->typePtr == &chanObjType) { /* + * TODO: TAINT Flag and dup'd channel values? * The channel is valid until any call to DetachChannel occurs. * Ensure consistency checks are done. */ statePtr = GET_CHANNELSTATE(objPtr); - interpPtr = GET_CHANNELINTERP(objPtr); if (GotFlag(statePtr, CHANNEL_TAINTED|CHANNEL_CLOSED)) { ResetFlag(statePtr, CHANNEL_TAINTED); Tcl_Release(statePtr); - UpdateStringOfChannel(objPtr); objPtr->typePtr = NULL; - } else if (interpPtr != (Interp*) interp) { + } else if (interp != GET_CHANNELINTERP(objPtr)) { Tcl_Release(statePtr); - UpdateStringOfChannel(objPtr); objPtr->typePtr = NULL; } } - if (objPtr->typePtr != &tclChannelType) { - Tcl_Channel chan; + if (objPtr->typePtr != &chanObjType) { + Tcl_Channel chan = Tcl_GetChannel(interp, TclGetString(objPtr), NULL); - /* - * We need a valid string with which to check for a valid channel, but - * make sure not to free internal rep until validated. [Bug 1847044] - */ - - if ((objPtr->typePtr != NULL) && (objPtr->bytes == NULL)) { - objPtr->typePtr->updateStringProc(objPtr); - } - - chan = Tcl_GetChannel(interp, objPtr->bytes, NULL); if (chan == NULL) { return TCL_ERROR; } @@ -11292,7 +10653,7 @@ SetChannelFromAny( Tcl_Preserve(statePtr); SET_CHANNELSTATE(objPtr, statePtr); SET_CHANNELINTERP(objPtr, interp); - objPtr->typePtr = &tclChannelType; + objPtr->typePtr = &chanObjType; } return TCL_OK; } @@ -11300,45 +10661,6 @@ SetChannelFromAny( /* *---------------------------------------------------------------------- * - * UpdateStringOfChannel -- - * - * Update the string representation for an object whose internal - * representation is "Channel". - * - * Results: - * None. - * - * Side effects: - * The object's string may be set by converting its Unicode represention - * to UTF format. - * - *---------------------------------------------------------------------- - */ - -static void -UpdateStringOfChannel( - Tcl_Obj *objPtr) /* Object with string rep to update. */ -{ - if (objPtr->bytes == NULL) { - ChannelState *statePtr = GET_CHANNELSTATE(objPtr); - const char *name = statePtr->channelName; - - if (name) { - size_t len = strlen(name); - - objPtr->bytes = ckalloc(len + 1); - objPtr->length = len; - memcpy(objPtr->bytes, name, len); - } else { - objPtr->bytes = tclEmptyStringRep; - objPtr->length = 0; - } - } -} - -/* - *---------------------------------------------------------------------- - * * FreeChannelIntRep -- * * Release statePtr storage. @@ -11381,20 +10703,14 @@ DumpFlags( ChanFlag('n', CHANNEL_NONBLOCKING); ChanFlag('l', CHANNEL_LINEBUFFERED); ChanFlag('u', CHANNEL_UNBUFFERED); - ChanFlag('R', BUFFER_READY); ChanFlag('F', BG_FLUSH_SCHEDULED); ChanFlag('c', CHANNEL_CLOSED); ChanFlag('E', CHANNEL_EOF); ChanFlag('S', CHANNEL_STICKY_EOF); ChanFlag('B', CHANNEL_BLOCKED); ChanFlag('/', INPUT_SAW_CR); - ChanFlag('*', INPUT_NEED_NL); ChanFlag('D', CHANNEL_DEAD); ChanFlag('R', CHANNEL_RAW_MODE); -#ifdef TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING - ChanFlag('T', CHANNEL_TIMER_FEV); - ChanFlag('H', CHANNEL_HAS_MORE_DATA); -#endif /* TCL_IO_TRACK_OS_FOR_DRIVER_WITH_BAD_BLOCKING */ ChanFlag('x', CHANNEL_INCLOSE); buf[i] ='\0'; |