| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
|
|
|
|
| |
to delete the async handler (common trap). As, only the one
that created the handler is allowed to delete it.
|
|
|
|
|
| |
to locate handler token fails. Happens when some other
thread attempts to delete somebody else's token.
|
|
|
|
|
|
|
|
|
|
| |
implementation. Added test cases about how it handles if the rug
is pulled out from under a channel (= killing threads,
interpreters containing the tcl command for a channel, and channel
sitting in a different interpreter/thread.)
* generic/tclIORChan.c: Fixed the bugs exposed by the new
testcases, redone most of the cleanup and exit handling.
|
|
|
|
|
|
|
| |
* io.test (io-53.8a): Ferrieux <ferrieux@users.sourceforge.net>,
* chanio.test (chan-io-53.8a): to shift EOF handling to the async
part of the command if a callback is specified, should the channel
be at EOF already when fcopy is called. Testcase by myself.
|
|
|
|
|
|
|
|
|
|
|
| |
* library/init.tcl: CVS development snapshots from the 8.5.2 and
* unix/configure.in: 8.5.3 releases.
* unix/tcl.spec:
* win/configure.in:
* README
* unix/configure: autoconf (2.59)
* win/configure:
|
|
|
|
|
|
|
|
|
|
|
|
| |
values, changed to not be an error, but behave like the special
value -1 (copy all, default).
* tests/iocmd.test (iocmd-15.{12,13}): Removed.
* tests/io.test (io-52.5{,a,b}): Reverted last change, added
* tests/chanio.test (chan-io-52.5{,a,b}): comment regarding the
meaning of -1, added two more testcases for other negative values,
and input wrapped to negative.
|
|
|
|
|
|
|
|
| |
* tests/ioCmd.test (iocmd-15.{13,14}): value to reject negative
values, and values overflowing 32-bit signed. [Bug 1557855]. Basic
patch by Alexandre Ferrieux <ferrieux@users.sourceforge.net>, with
modifications from me to separate overflow from true negative
value. Extended testsuite.
|
| |
|
|
|
|
|
|
|
|
| |
* tests/chanio.test:
* generic/tclIO.c: Additional changes to data structures for fcopy
* generic/tclIO.h: and channels to perform proper cleanup in case
of a channel having two background copy operations running as is
now possible.
|
|
|
|
|
|
| |
TclCopyChannel): New macro, and the places using it. This change
allows for bi-directional fcopy on channels. Thanks to Alexandre
Ferrieux <ferrieux@users.sourceforge.net> for the patch.
|
|
|
|
|
|
| |
Fix [format {% d}] so that it behaves the same way as in 8.4 and
as C's printf().
* tests/format.test: Add a test for '% d' and '%+d'.
|
|
|
|
|
|
|
|
| |
write when input was zero-length anyway. Otherwise keept it an
error, and separate the message from 'written too much'.
* tests/ioCmd.test (iocmd-24.6): Testcase updated for changed
message.
|
|
|
|
|
|
|
|
|
| |
the now closed channel from the reflection map. Before we could
crash the system by invoking 'chan postevent' on a closed
reflected channel, dereferencing the dangling pointer in the
map.
* tests/ioCmd.test (iocmd-31.8): Testcase for the above.
|
|
|
|
|
|
| |
* tests/io.test: prevent fcopy from calling -command synchronously
* tests/chanio.test: the first time. Thanks to Alexandre Ferrieux
<ferrieux@users.sourceforge.net> for report and patch.
|
|
|
|
|
|
|
| |
[Bug 780533], with many thanks to Alexandre Ferrieux
<ferrieux@users.sourceforge.net> for tracking it down and
providing a solution. Still have to convert his test script into
a proper test case.
|
|
|
|
|
| |
* unix/tcl.m4: rounding setup on solaris x86, native cc), provided
* unix/configure: by Michael Schlenker. configure regen'd.
|
|
|
|
| |
* generic/tclPkg.c (Tcl_PkgInitStubsCheck):
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
* library/init.tcl:
* tools/tcl.wse.in:
* unix/configure.in:
* unix/tcl.spec:
* win/configure.in:
* unix/configure: autoconf-2.59
* win/configure:
|
| |
|
|
|
|
| |
condition.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
alignment of memory allocated by GrowEvaluationStack(). [Bug 1914503]
|
|
|
|
| |
confusing the simple-minded code doing the calculations). [Bug 1903325]
|
|
|
|
| |
from bad TclStackAlloc() alignment. Thanks George Peter Staplin.
|
|
|
|
|
| |
SortElement arrays instead of TclStackAlloc() which isn't getting
alignment right. Workaround for [Bug 1914503].
|
|
|
|
| |
* unix/tclUnixPipe.c: we are about to exit anyways.
|
| |
|
|
|
|
|
| |
with libm in Cygwin and DJGPP. Thanks to Gordon Schumacher and
Philip Moore. [Patch 1800636]
|
|
|
|
|
| |
sub-encodings, fixes encoding-11.1 failing after iso2022-jp loaded
(e.g. TESTFLAGS='-singleproc 1 -match "chan-io-1.8 encoding-11.1"').
|
|
|
|
|
|
|
| |
* tests/var.test (var-8.2): unset traces on vars should be called
with a FQ named during namespace deletion. This was causing
infinite loops when unset traces recreated the var, as reported by
Julian Noble on [Bug 1911919].
|
|
|
|
|
| |
* tests/event.test (event-5.*): checking to protect against callers
passing invalid return options dictionaries. [Bug 1901113]
|
|
|
|
|
|
| |
* tests/expr.test: function and the [::tcl::mathfunc::abs]
command do not return the value of -0, or equivalent values with
more alarming string reps like -1e-350. [Bug 1893815].
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
1904907]. Reset the return option keys to NULL to allow full
re-initialization by GetKeys(). This introduces a memory leak
for the key objects, but gets us around a crash in the
finalization of reflected channels when handling returns, either
at compile- or runtime. In both cases we access the keys after
they have been released by their thread exit handler. A proper
fix is entangled with the untangling of the finalization
ordering and attendant issues. For now we choose the lesser
evil.
|
| |
|
|
|
|
|
|
| |
compiling so that bytecodes invalid due to changing context or due
to the difference between expressions and scripts are not reused.
[Bug 1899164].
|
|
|
|
|
| |
[expr] so that [expr $e] caches compiled bytecodes for the expression
as the intrep of $e.
|
|
|
|
|
| |
an nsNameType obj when the reference crosses interpreter
boundaries.
|
|
|
|
|
|
|
|
|
|
| |
management of Tcl_SetReturnOptions to become that of a conventional
Consumer routine. Thanks to Peter Spjuth for pointing out the
difficulties calling Tcl_SetReturnOptions with non-0-count value for
options.
* generic/tclExecute.c (INST_RETURN_STK): Revised the one caller
within Tcl itself which passes a non-0-count value to
Tcl_SetReturnOptions().
|
|
|
|
|
|
|
|
|
|
| |
refcount management of Tcl_AppendObjToErrorInfo to become that of
a conventional Consumer routine. This preserves the ease of use
for the overwhelming common callers who pass in a 0-count value,
but makes the proper call with a non-0-count value less surprising.
* generic/tclEvent.c (TclDefaultBgErrorHandlerObjCmd): Revised the
one caller within Tcl itself which passes a non-0-count value to
Tcl_AppendObjToErrorInfo().
|
|
|
|
|
|
| |
an enter trace deletes or changes the command, prompting a reparsing.
Don't let the second pass lose commandPtr value allocated during the
first pass.
|
|
|
|
| |
message generation.
|