| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* generic/tclCmdIL.c:
* generic/tclCompile.c:
* generic/tclExecute.c:
* generic/tclHash.c:
* generic/tclIOUtil.c:
* generic/tclVar.c:
* generic/tclBasic.c: Silence compiler warnings about ClientData.
* generic/tclProc.c:
* generic/tclScan.c: Typo in ACCEPT_NAN configuration.
* generic/tclStrToD.c: Set floating point control register on
MIPS systems so that the gradual underflow expected by Tcl is
in effect. [Bug 2819200]
|
|
|
|
| |
Also added code from itcl-ng for better separation of concerns.
|
| |
|
|
|
|
|
|
|
| |
* generic/tclBasic.c (builtInCmds):
* generic/tclCmdMZ.c (Tcl_SwitchObjCmd):
* tests/switch.test (switch-15.1):
Make non-bytecoded [switch] command aware of NRE. [Bug 2821401]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
TclCleanupByteCode, TclCompileScript):
* generic/tclExecute.c (TclCompileObj, TclExecuteByteCode):
* tclCompile.h (ExtCmdLoc):
* tclInt.h (ExtIndex, CFWordBC, CmdFrame):
* tclBasic.c (DeleteInterpProc, TclArgumentBCEnter,
TclArgumentBCRelease, TclArgumentGet, SAVE_CONTEXT,
RESTORE_CONTEXT, NRCoroutineExitCallback, TclNRCoroutineObjCmd):
* generic/tclCmdAH.c (TclNRForObjCmd, TclNRForIterCallback,
ForNextCallback):
* generic/tclCmdMZ.c (TclNRWhileObjCmd):
Extended the bytecode compiler initialization to recognize the
compilation of whole files (NRE enabled 'source' command) and
switch to the counting of absolute lines in that case.
Further extended the bytecode compiler to track the start line in
the generated information, and modified the bytecode execution to
recompile an object if the location as per the calling context
doesn't match the location saved in the bytecode. This part could
be optimized more by using more memory to keep all possibilities
which occur around, or by just adjusting the location information
instead of a total recompile.
Reworked the handling of literal command arguments in bytecode to
be saved (compiler) and used (execution) per command (See the
TCL_INVOKE_STK* instructions), and not per the whole bytecode.
This, and the previous change remove the problems with location
data caused by literal sharing (across whole files, but also proc
bodies). Simplified the associated datastructures (ExtIndex is
gone, as is the function EnterCmdWordIndex).
The last change causes the hashtable 'lineLABCPtr' to be state
which has to be kept per coroutine, like the CmdFrame stack.
Reworked the coroutine support code to create, delete and switch
the information as needed. Further reworked the tailcall command
as well, it has to pop its own arguments when run in a bytecode
context to keep a proper stack in 'lineLABCPtr'.
Fixed the mishandling of line information in the NRE-enabled 'for'
and 'while' commands introduced when both were made to share their
iteration callbacks without taking into account that the loop body
is found in different words of the command. Introduced a separate
data structure to hold all the callback information, as we went
over the limit of 4 direct client-data values for NRE callbacks.
The above fixes [Bug 1605269].
|
| |
|
|
|
|
| |
[Bug 1895546]
|
| |
|
|
|
|
|
| |
Tcl_SetChannelError requires a list. Added some tests to ensure error propagation
from the zlib library to the interp.
|
|
|
|
| |
format.
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
| |
* generic/tclPanic.c: analyzer and redefine Tcl_Panic to
* generic/tclStubInit.c: assert after panic in clang PURIFY
builds.
* generic/tclCmdIL.c: add clang assert for false positive
from static analyzer.
|
| |
|
| |
|
|
|
|
| |
[exit] wrt finalization, and thus solving many multi-thread teardown issues [Bugs 2001201, 486399, and possibly 597575, 990457, 1437595, 2750491].
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* generic/tclProc.c: when compiling a proc survives too long. We
* tests/execute.test: only need it there long enough for the right
TclInitCompileEnv() call to re-stash it into envPtr->procPtr. Once
that is done, the CompileEnv controls. If we let the value of
iPtr->compiledProcPtr linger, though, then any other bytecode compile
operation that takes place will also have its CompileEnv initialized
with it, and that's not correct. The value is meant to control the
compile of the proc body only, not other compile tasks that happen
along. Thanks to Carlos Tasada for discovering and reporting the
problem. [Bug 2802881].
|
|
|
|
|
|
|
| |
integer calculations computing the length of the %ll formats of
really big integers. Also added protections so that [format]s that
would produce results overflowing the maximum string length of Tcl
values throw a normal Tcl error instead of a panic. [Bug 2801413]
|
|
|
|
|
| |
"pure unicode" representation of an empty string. Thanks to Julian
Noble for reporting the problem. [Bug 2803109]
|
|
|
|
|
|
|
|
|
|
| |
* library/clock.tcl: Bison parser (needed a %pure-parser
* tests/clock.test: declaration to avoid static variables).
Discovered that the %pure-parser declaration
allowed for returning the Bison error message
to the Tcl caller in the event of a syntax
error, so did so.
* generic/tclDate.c: bison 2.3
|
| |
|
|
|
|
|
| |
a table of static constants in the lookup table for exponent operator
computations that fit in a 64 bit integer result.
|
|
|
|
| |
logic of the INST_EXPON instruction to fix [Bug 2798543].
|
| |
|
| |
|
| |
|
|
|
|
|
| |
insure that a command in a deleted namespace cannot be found
through a cached name.
|
|
|
|
|
| |
* generic/tclCompile.h: stack: 200 words (previously was 2000,
* generic/tclExecute.c: the same as interps)
|
|
|
|
| |
[string repeat]. [Bug 2582327]
|
| |
|
| |
|
|
|
|
| |
is not used, use Tcl_NRCallObjProc instead [Bug 2486550].
|
| |
|
| |
|
|
|
|
|
| |
flag inside the Tcl_WrongNumArgs function,
so the caller no longer has to do the reset.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* unix/tclUnixChan.c: embedding into applications that
* unix/tclUnixEvent.c: already have a CFRunLoop running and
want to run the tcl event loop via
Tcl_ServiceModeHook(TCL_SERVICE_ALL).
* macosx/tclMacOSXNotify.c: add CFRunLoop based Tcl_Sleep() and
* unix/tclUnixChan.c: TclUnixWaitForFile() implementations
* unix/tclUnixEvent.c: and disable select() based ones in
CoreFoundation builds.
* unix/tclUnixNotify.c: simplify, sync with tclMacOSXNotify.c.
* generic/tclInt.decls: add TclMacOSXNotifierAddRunLoopMode()
* generic/tclIntPlatDecls.h: internal API, regen.
* generic/tclStubInit.c:
|
|
|
|
| |
TCL_GROWTH_MIN_ALLOC is everywhere expressed in bytes as comment claims.
|
|
|
|
|
|
| |
* tests/fileName.test: the wrong results for both [file dirname] and
[file tail] on "path" arguments with the PATHFLAGS != 0 intrep and
with an empty string for the "joined-on" part. [Bug 2710920]
|
| |
|
| |
|
|
|
|
| |
[Bug 2502037].
|
|
|
|
|
|
|
| |
* generic/tclInt.h: ensembles. Only remaining known leak
* generic/tclInterp.c: is in ensemble unknown dispatch (as it
* generic/tclNamesp.c: not NR-enabled)
* tests/tailcall.test:
|
|
|
|
|
|
| |
* tests/tailcall.test: added tests to show that [tailcall] does
not currently always execute in constant space: interp-alias,
ns-imports and ensembles "leak" as of this commit.
|
|
|
|
|
|
|
|
|
| |
* generic/tclCmdAH.c: Tailcalls now perform properly even from
* generic/tclExecute.c: within [eval]ed scripts.
* generic/tclInt.h: More tests missing, as well as proper
exploration and testing of the interaction with "redirectors" like
interp-alias (suspect that it does not happen in constant space)
and pure-eval commands.
|