| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
| |
MACOSX_DEPLOYMENT_TARGET unset. [Bug 1543715]
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
already running (e.g. if Tcl_WaitForEvent was called recursively),
re-run it in a custom run loop mode containing only the source for the
notifier thread, otherwise wakeups from other sources added to the
common run loop modes might get lost; sync panic msg changes from HEAD.
* unix/tclUnixNotfy.c (Tcl_WaitForEvent): on 64-bit Darwin,
pthread_cond_timedwait() appears to have a bug that causes it to wait
forever when passed an absolute time which has already been exceeded by
the system time; as a workaround, when given a very brief timeout, just
do a poll on that platform. [Bug 1457797]
|
|
|
|
|
| |
returns early (e.g. due to a signal), call it again instead of returning
a timeout result. Fixes intermittent event-13.8 failures.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
universal builds including x86_64, for 64-bit CoreFoundation on Leopard
and for use of -mmacosx-version-min instead of MACOSX_DEPLOYMENT_TARGET.
* unix/configure: autoconf-2.13
* generic/tcl.h: add fixes for building on Leopard and support for
* unix/tclUnixPort.h: 64-bit CoreFoundation on Leopard.
* unix/tclUnixPort.h: on Darwin x86_64, disable use of vfork as it
causes execve to fail intermittently. (rdar://4685553)
* macosx/README: updates for x86_64 support and Xcode 2.3.
|
|
|
|
| |
with call to memcpy(). Thanks to afredd. [Patch 1530262]
|
|
|
|
|
| |
channels are initialized before vfork() so that the child doesn't
potentially corrupt global state in the parent's address space.
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
notifier thread lazily upon first call to Tcl_WaitForEvent() rather than
in Tcl_InitNotifier(). Allows calling exeve() in processes where the
event loop has not yet been run (Darwin's execve() fails in processes
with more than one thread), in particular allows embedders to call
fork() followed by execve(), previously the pthread_atfork() child
handler's call to Tcl_InitNotifier() would immediately recreate the
notifier thread in the child after a fork.
* macosx/tclMacOSXNotify.c (Tcl_InitNotifier): add support for
* unix/tclUnixFCmd.c (DoRenameFile, CopyFileAtts): weakly importing
* unix/tclUnixInit.c (TclpSetInitialEncodings): symbols not available
on OSX 10.2 or 10.3, enables binaires built on later OSX versions to run
on earlier ones.
* macosx/README: document how to enable weak-linking; cleanup.
* unix/tclUnixPort.h: add support for weak-linking; conditionalize
AvailabilityMacros.h inclusion; only disable realpath on 10.2 or earlier
when threads are enabled.
* unix/tclLoadDyld.c (TclpLoadMemoryGetBuffer): change runtime Darwin
* unix/tclUnixInit.c (TclpInitPlatform): release check to use
global initialized once.
* unix/tclUnixFCmd.c (DoRenameFile, TclpObjNormalizePath): add runtime
Darwin release check to determine if realpath is threadsafe.
* unix/configure.in: add check on Darwin for compiler support of weak
* unix/tcl.m4: import and for AvailabilityMacros.h header; move
Darwin specific checks & defines that are only relevant to the tcl build
out of tcl.m4; restrict framework option to Darwin; cleanup quoting.
* unix/configure: autoconf-2.13
* unix/tclLoadDyld.c (TclpLoadMemory):
* unix/tclUnixPipe.c (TclpCreateProcess): fix signed-with-unsigned
comparison and other warnings from gcc4 -Wextra.
|
|
|
|
|
|
| |
<AvailabilityMacros.h>. The missing header caused the upcoming
#if conditions to wrongly exclude realpath, causing file
normalize to ignore symbolic links in the path.
|
|
|
|
| |
has been garbage-collected.
|
|
|
|
| |
inheritance to be #defined out (default remains in).
|
| |
|
| |
|
| |
|
|
|
|
|
| |
define from AvailabilityMacros.h: override configure detection and only
use API available in the indicated OS version or earlier.
|
| |
|
|
|
|
|
|
| |
(VS2005). We have to handle a number of oddities as they have
deprecated most of the standard C library and now generate manifest
files to be linked into the binaries. SF bug #1424909
|
| |
|
|
|
|
| |
Tcl_SetStdChannel() be accompanied by a call to Tcl_RegisterChannel().
|
|
|
|
|
|
| |
fix of 2006-01-09 for [namespace inscope] as it seems to mess with
itcl scope decoding. Leaving namespace-29.6 test failure until
final cause it determined.
|
|
|
|
| |
placement in unbraced outer if/else conditions. (jcw)
|
|
|
|
|
|
|
|
|
|
|
|
| |
* unix/tcl.m4 (Darwin): recreates CoreFoundation state and notifier
thread in the child after a fork(). Note that pthread_atfork() is
available starting with Tiger only. Because vfork() is used by the core
on Darwin, [exec]/[open] are not affected by this fix, only extensions
or embedders that call fork() directly (such as TclX). However, this
only makes fork() safe from corefoundation tcl with --disable-threads;
as on all platforms, forked children may deadlock in threaded tcl due to
the potential for stale locked mutexes in the child. [Patch 923072]
* unix/configure: autoconf-2.59
|
| |
|
| |
|
| |
|
|
|
|
|
|
| |
normalization done by the recursive glob routine, since changing the
precise string of the pathname broke [glob] on some Tcl_Filesystems.
[Bug 943995]
|
|
|
|
|
|
|
|
| |
epoch forces the re-compile of a proc body, take care not to
overwrite any Proc struct that may be referred to on the active
call stack. This fixes [Bug 148218]. Note that the fix will not be
effective for code that calls the private routine TclProcCompileProc()
directly.
|
|
|
|
|
|
| |
* tests/main.test: (Tcl_Main-4.5): processing of one interactive
command before passing control to the loop routine registered with
Tcl_SetMainLoop() [Bug 1481986].
|
|
|
|
|
|
|
|
|
|
|
|
| |
* generic/tcl.h:
* tools/tcl.wse.in:
* unix/configure.in:
* unix/tcl.spec:
* win/README.binary:
* win/configure.in:
* unix/configure: autoconf-2.13
* win/configure:
|
|
|
|
| |
* tests/expr-old.test: integer values by srand() [Bug 1480509].
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
| |
* tests/trace.test: and enterstep traces. [Bug 1458266]
|
|
|
|
| |
* changes: updates for another RC.
|
|
|
|
| |
indicate that readiness for reinitialization.
|
|
|
|
|
|
|
| |
* tests/indexObj.test: there are extensions that rely on the prior
* doc/GetIndex.3: behavior that the empty string cannot succeed
as a unique prefix matcher, so I'm restoring Donal Fellow's solution.
Added mention of this detail to the documentation. [Bug 1464039]
|
|
|
|
|
| |
define on Darwin. [Bug 1457515]
* unix/configure: autoconf-2.13
|
| |
|
|
|
|
|
|
|
| |
* win/tclWinDde.c:
* win/Makefile.in:
* win/configure.in:
* win/configure: autoconf 2.13
|
|
|
|
|
|
| |
package * win/tclWinReg.c: version 1.1.4 (should have been done
for the Tcl 8.4.8 release!)
|
|
|
|
|
| |
* win/tclWinDde.c: version 1.2.4 (should have been done
for the Tcl 8.4.8 release!)
|
|
|
|
|
| |
strings to be matched by the Tcl_GetIndexFromObj machinery, in
the same manner as any other key. [Bug 1464039]
|