| Ticket UUID: | 853643 | |||
| Title: | Index script preference for debug Tk | |||
| Type: | Patch | Version: | None | |
| Submitter: | dgp | Created on: | 2003-12-03 22:28:50 | |
| Subsystem: | 85. Win Build | Assigned To: | davygrvy | |
| Priority: | 5 Medium | Severity: | ||
| Status: | Closed | Last Modified: | 2026-10-22 17:22:37 | |
| Resolution: | Out of Date | Closed By: | serhiy.storchaka | |
| Closed on: | 2026-10-04 07:57:48 | |||
| Description: |
When both debug and non-debug Tk are installed, the generated index scripts allow either one to successfully [load] into tclsh. Which one you get from a [package require Tk] is currently quasi-random, depending on the order in which [tclPkgUnknown] [source]s the pkgIndex.tcl files. This patch forces a preference for the debug-enabled Tk when Tcl is also debug-enabled. David Gravereaux reports he finds this combination useful when stepping through Tcl/Tk applications. When Tcl is not debug-enabled, the choice for Tk remains quasi-random even after this patch, but that seems OK. Either Tk will work, so either satisfies the [package require]. This patch changes only win/Makefile. If it is accepted, similar changes should be made to the alternative makefiles. | |||
| User Comments: |
serhiy.storchaka added on 2026-10-04 07:57:48:
Out of date: TK_DBGX no longer exists. dgp added on 2004-12-10 01:41:44: Logged In: YES user_id=80530 this will be moot once Patch 1081595 kills TCL_DBGX, right? davygrvy added on 2003-12-24 09:43:56: File Added - 71396: pkgIndex.tcl.txt Logged In: YES user_id=7549 Here's something I'm doing for Itcl. The Stubbed dll can load into any core 8.1+ and there's also an 8.0 one, too. What do you think? Tk needs to match itself to the core version. dgp added on 2003-12-18 06:13:57: Logged In: YES user_id=80530 weird, apparently on Unix, Tk's configure doesn't honor the --enable-threads option, but takes whatever was set in the Tcl build from the corresponding tclConfig.sh. Not clear what effect that has on [load]-ability into a different Tcl. (binary extension distribution anyone?) On Windows, though, looks like it does respect the --enable-threads choice. Better take this to e-mail or chat, or just individual study. dgp added on 2003-12-18 05:55:58: Logged In: YES user_id=80530 ok, I slowly see more. If Tk ends up paired with a --disable-threads build of Tcl, then all these routines are no-ops anyway, so using macros to remove the calls completely changes nothing and saves some time, and that's what --disable-threads build of Tk does. So the --enable-threads and --disable-threads builds are different. Either Tk build will [load] into either Tcl variant, it appears, but the only way the combination will support threads is if both are --enable-threads configurations. davygrvy added on 2003-12-18 05:44:56: Logged In: YES user_id=7549 Pretty close. Tcl_Mutex(un)lock are macros and empty without #define TCL_THREADS for the Tk build. I don't know off-hand about the TCL_TSD_INIT macro. It would be nice if the TCL_TSD_INIT macro moved to tcl.h from tclInt.h, but that's another issue and very minor. dgp added on 2003-12-18 05:19:35: Logged In: YES user_id=80530 It looks to me like those public Tcl routines are all in the public stubs table, so the Tk library, built with -DUSE_TCL_STUBS is going to look the same with either configuration, isn't it? In other words, if tk85t.dll and tk85.dll are byte for byte identical, I don't see the benefit in having both names (or even recognizing a --enable-threads configuration option for that matter). As far as I'm concerned the fewer variants to keep track of, the better. I'll take a brief look at the Tk code myself to see if I can find anywhere that a --enable-threads configuration actually has an effect. davygrvy added on 2003-12-18 03:20:22: Logged In: YES user_id=7549 Tcl_GetThreadData, Tcl_MutexLock, Tcl_MutexUnlock and a few other bits actually have an implimentation when the core is threaded. That's about it. When not threaded, Tcl_GetThreadData just makes or gets an instance of the ThreadSpecificData struct but isn't associated to the calling thread (just one). Again, behavior of threading is determined by the core the extension is loaded into. If a threaded Tk loads into a non- threaded Tcl, no thread specific data is maintained and mutexes don't work. As far as I know, this is absolute.. I don't know if there are any exceptions. dgp added on 2003-12-18 02:30:53: Logged In: YES user_id=80530 What is the difference between a --enable-threads build of Tk and a --disable-threads build of Tk. Are the DLLs produced really different? If so, is there anything one can do that the other cannot? Ideally, I think I'd like Tk not to care about threads. Short of that I'd like a better understanding about why it does care. davygrvy added on 2003-12-18 02:06:27: Logged In: YES user_id=7549 An order of preference might be: if Tcl is not threaded and a threaded Tk exists, prefer the threaded Tk. The behavior of the TSD and mutexes is determined by the Tcl core the extension is loaded into. IMO, all extensions should be compiled for thread support by default. If Tcl has symbols, prefer a symbols Tk, but fallback to non- symbols when not available. memdbg and compstats are a grey area. The prefix is the same for symbols. 'g' has something regarding debugging, but can not be determined. davygrvy added on 2003-12-18 01:49:59: Logged In: YES user_id=7549 C:\PROGRA~1\Tcl\bin>dir /b wish85* wish85.exe wish85g.exe wish85t.exe wish85tg.exe C:\PROGRA~1\Tcl\bin>dir /b tk85* tk85.dll tk85g.dll tk85t.dll tk85tg.dll The g one is straight symbols without memdbg nor compstats, but we wouldn't know this from the suffix. I only have one lib/tk8.5 directory. dgp added on 2003-12-18 00:56:58: Logged In: YES user_id=80530 Essentially yes. tk/win/Makefile installs whatever variant was built into $PKG_INSTALL_DIR which is defined as $(LIB_INSTALL_DIR)/tk$(VERSION)$(TK_DBGX) so there as many distinct install directories as there are variations of $TK_DBGX. For variants not distinguished by TK_DBGX, let's discuss further. Which of these variants does it make sense to have concurrently installed? What distinct compatibility needs do they have with each other, and with Tcl and other libraries? See also: http://wiki.tcl.tk/4312 davygrvy added on 2003-12-17 07:12:06: Logged In: YES user_id=7549 Is this patch saying alternate build flavors of Tk should install to a different directory? As in: lib/tk8.5 lib/tk8.5dbg lib/tk8.5thrd lib/tk8.5compstat lib/tk8.5dbgthrd lib/tk8.5memdbg lib/tk8.5memdbgthrd ..etc, etc.. dgp added on 2003-12-04 05:28:50: File Added - 69481: debug.patch | |||
Attachments:
- pkgIndex.tcl.txt [download] added by davygrvy on 2003-12-24 09:43:56. [details]
- debug.patch [download] added by dgp on 2003-12-04 05:28:50. [details]
