cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn #18

jeremyd2019 · 2020-11-08T19:19:27Z

This allows subprocesses to get Windows-default error handling behavior (such as invoking the registered JIT debugger).

Fixes https://sourceforge.net/p/msys2/discussion/general/thread/1daf3345/ (google turned this thread up when I was looking up how to use error_start)

dscho · 2020-11-08T23:38:08Z

* should this be behind an option (like in the CYGWIN/MSYS2 environment variable)?

I think that would be a good idea (note, though, that it is the MSYS variable, not the MSYS2 variable).

The reason I say this is that I would like to have as vanilla an MSYS2 runtime as possible in Git for Windows, where we definitely do not want to invoke the registered debugger.

FWIW it is a bit unclear to me how useful this behavior is in practice: I only know of the Visual Studio debugger as being an option for this, and we do not have any .pdb files for Visual Studio to work with.

If it turns out to be quite useful, I would strongly recommend sending this patch to the Cygwin project, too.

jeremyd2019 · 2020-11-08T23:58:48Z

FWIW it is a bit unclear to me how useful this behavior is in practice: I only know of the Visual Studio debugger as being an option for this, and we do not have any .pdb files for Visual Studio to work with.

I actually use WinDbg/WinDbgX, now that I've discovered cv2pdb. But it is possible to use GDB in this way too: https://www.sourceware.org/gdb/current/onlinedocs/gdb/Cygwin-Native.html (not sure why this is documented in a file called 'Cygwin-Native')

To use it, create or edit the following keys in HKLM\SOFTWARE\Microsoft\Windows NT\CurrentVersion\AeDebug and/or HKLM\SOFTWARE\Wow6432Node\Microsoft\Windows NT\CurrentVersion\AeDebug (for x86_64 versions):

Debugger (REG_SZ) — a command to launch the debugger. Suggested command is: fully-qualified-path-to-gdb.exe -ex "attach %ld" -ex "signal-event %ld" -ex "continue".
The first %ld will be replaced by the process ID of the crashing process, the second %ld will be replaced by the ID of the event that blocks the crashing process, waiting for GDB to attach.

Auto (REG_SZ) — either 1 or 0. 1 will make the system run debugger specified by the Debugger key automatically, 0 will cause a dialog box with “OK” and “Cancel” buttons to appear, which allows the user to either terminate the crashing process (OK) or debug it (Cancel).

jeremyd2019 · 2020-11-09T00:30:57Z

* should this be behind an option (like in the CYGWIN/MSYS2 environment variable)?
I think that would be a good idea (note, though, that it is the MSYS variable, not the MSYS2 variable).

Great, now I have to come up with a name for this 😬

jeremyd2019 · 2020-11-09T01:21:14Z

Best I could come up with for a name was spawndflterrmode but I am definitely open to suggestions. The more I look at that the less I like it 🤮

dscho · 2020-11-10T14:30:54Z

Why not jitdebugging?

dscho · 2020-11-10T14:31:41Z

it is possible to use GDB in this way too: https://www.sourceware.org/gdb/current/onlinedocs/gdb/Cygwin-Native.html (not sure why this is documented in a file called 'Cygwin-Native')

With that as supporting evidence, it should be easy to "sell" this patch upstream, too.

jeremyd2019 · 2020-11-10T18:22:21Z

Why not jitdebugging?

Well, cygwin processes will still require using error_start. Maybe something like winjitdebug. Though the SEM_NOGPFAULTERRORBOX flag also seems to suppress Windows Error Reporting. Unfortunate that so much was 'bolted on' to a flag that was only supposed to suppress a 'General Protection Fault' message on Windows 3.1.

mingwandroid · 2020-11-11T00:03:42Z

I used to use qtcreator as the debugger here when msys2 software crashed along with an msys2 kit in the ide. It worked really well in 2015 odd!

jeremyd2019 · 2020-11-11T19:53:20Z

I also think it would be useful to document how to do just-in-time debugging (such as in a 'wiki' page). For now, I've started a gist with what I've learned: https://gist.github.com/jeremyd2019/3069ae1d44e86f4695087b98271af5d9

dscho · 2020-12-03T08:24:57Z

I would strongly recommend sending this patch to the Cygwin project, too.

FWIW I still believe that would make sense.

jeremyd2019 · 2020-12-03T21:27:53Z

I would strongly recommend sending this patch to the Cygwin project, too.

FWIW I still believe that would make sense.

Just about forgot about this, since I finished with the crash I was trying to catch...
https://cygwin.com/pipermail/cygwin-patches/2020q4/010870.html

This allows native processes to get Windows-default error handling behavior (such as invoking the registered JIT debugger).

jeremyd2019 · 2020-12-10T18:35:21Z

Change was applied upstream after some discussion and adjustment: 21ec498

Upstream decided against a new option, and to apply the flag to all non-cygwin processes.

I'm not in a hurry for this change ATM, I'd be content to wait for it to show up in the next stable release, but I cherry-picked the commit and force-pushed prior to closing just in case.

dscho · 2021-03-31T08:53:43Z

Upstream decided against a new option, and to apply the flag to all non-cygwin processes.

Uh oh. That's really not good, and we only found out now: msys2/MSYS2-packages#2414 (comment)

In msys2/msys2-runtime#18, we discussed a change that would allow default Windows error handling of spawned processes to kick in (such as registered JIT debuggers). We even agreed that it would make sense to hide this functionality behind a flag, `winjitdebug`. However, when this got upstreamed as 21ec498 (cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn, 2020-12-09), that flag was deemed unnecessary. But it would appear that it _is_ necessary: As reported in msys2/MSYS2-packages#2414 (comment) this new behavior is pretty disruptive e.g. in CI scenarios. So let's introduce that `winjitdebug` flag (settable via the environment variable `MSYS`) at long last. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>

In msys2#18, we discussed a change that would allow default Windows error handling of spawned processes to kick in (such as registered JIT debuggers). We even agreed that it would make sense to hide this functionality behind a flag, `winjitdebug`. However, when this got upstreamed as 21ec498 (cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn, 2020-12-09), that flag was deemed unnecessary. But it would appear that it _is_ necessary: As reported in msys2/MSYS2-packages#2414 (comment) this new behavior is pretty disruptive e.g. in CI scenarios. So let's introduce that `winjitdebug` flag (settable via the environment variable `MSYS`) at long last. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>

jeremyd2019 · 2021-03-31T17:02:12Z

🤦 . I had been using this patch for a while, building clang packages, and I think I saw a debugger pop up once during configure, but no missing dll message boxes.

dscho · 2021-03-31T22:08:24Z

You probably did not try to run any .exe with a missing .dll ;-)

In #18, we discussed a change that would allow default Windows error handling of spawned processes to kick in (such as registered JIT debuggers). We even agreed that it would make sense to hide this functionality behind a flag, `winjitdebug`. However, when this got upstreamed as 21ec498 (cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn, 2020-12-09), that flag was deemed unnecessary. But it would appear that it _is_ necessary: As reported in msys2/MSYS2-packages#2414 (comment) this new behavior is pretty disruptive e.g. in CI scenarios. So let's introduce that `winjitdebug` flag (settable via the environment variable `MSYS`) at long last. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>

In msys2/msys2-runtime#18, we discussed a change that would allow default Windows error handling of spawned processes to kick in (such as registered JIT debuggers). We even agreed that it would make sense to hide this functionality behind a flag, `winjitdebug`. However, when this got upstreamed as 21ec498 (cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn, 2020-12-09), that flag was deemed unnecessary. But it would appear that it _is_ necessary: As reported in msys2/MSYS2-packages#2414 (comment) this new behavior is pretty disruptive e.g. in CI scenarios. So let's introduce that `winjitdebug` flag (settable via the environment variable `MSYS`) at long last. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>

In #18, we discussed a change that would allow default Windows error handling of spawned processes to kick in (such as registered JIT debuggers). We even agreed that it would make sense to hide this functionality behind a flag, `winjitdebug`. However, when this got upstreamed as 21ec498 (cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn, 2020-12-09), that flag was deemed unnecessary. But it would appear that it _is_ necessary: As reported in msys2/MSYS2-packages#2414 (comment) this new behavior is pretty disruptive e.g. in CI scenarios. So let's introduce that `winjitdebug` flag (settable via the environment variable `MSYS`) at long last. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>

In msys2/msys2-runtime#18, we discussed a change that would allow default Windows error handling of spawned processes to kick in (such as registered JIT debuggers). We even agreed that it would make sense to hide this functionality behind a flag, `winjitdebug`. However, when this got upstreamed as 21ec498 (cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn, 2020-12-09), that flag was deemed unnecessary. But it would appear that it _is_ necessary: As reported in msys2/MSYS2-packages#2414 (comment) this new behavior is pretty disruptive e.g. in CI scenarios. So let's introduce that `winjitdebug` flag (settable via the environment variable `MSYS`) at long last. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>

In #18, we discussed a change that would allow default Windows error handling of spawned processes to kick in (such as registered JIT debuggers). We even agreed that it would make sense to hide this functionality behind a flag, `winjitdebug`. However, when this got upstreamed as 21ec498 (cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn, 2020-12-09), that flag was deemed unnecessary. But it would appear that it _is_ necessary: As reported in msys2/MSYS2-packages#2414 (comment) this new behavior is pretty disruptive e.g. in CI scenarios. So let's introduce that `winjitdebug` flag (settable via the environment variable `MSYS`) at long last. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>

...in a non-Cygwin child process. Backported from MSYS2. Downstream commit message follows. In msys2/msys2-runtime#18, we discussed a change that would allow default Windows error handling of spawned processes to kick in (such as registered JIT debuggers). We even agreed that it would make sense to hide this functionality behind a flag, `winjitdebug`. However, when this got upstreamed as 21ec498 (cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn, 2020-12-09), that flag was deemed unnecessary. But it would appear that it _is_ necessary: As reported in msys2/MSYS2-packages#2414 (comment) this new behavior is pretty disruptive e.g. in CI scenarios. So let's introduce that `winjitdebug` flag (settable via the environment variable `MSYS`) at long last. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Corinna Vinschen <corinna@vinschen.de>

In #18, we discussed a change that would allow default Windows error handling of spawned processes to kick in (such as registered JIT debuggers). We even agreed that it would make sense to hide this functionality behind a flag, `winjitdebug`. However, when this got upstreamed as 21ec498 (cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn, 2020-12-09), that flag was deemed unnecessary. But it would appear that it _is_ necessary: As reported in msys2/MSYS2-packages#2414 (comment) this new behavior is pretty disruptive e.g. in CI scenarios. So let's introduce that `winjitdebug` flag (settable via the environment variable `MSYS`) at long last. Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>

jeremyd2019 force-pushed the msys2-3_1_7-spawn-default-errormode branch from c48d6a7 to c5a53a7 Compare November 9, 2020 01:19

jeremyd2019 force-pushed the msys2-3_1_7-spawn-default-errormode branch from c5a53a7 to fdf1a51 Compare November 10, 2020 18:49

jeremyd2019 changed the title ~~[RFC] cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn~~ cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn Nov 10, 2020

jeremyd2019 marked this pull request as ready for review November 10, 2020 19:29

cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn

496311d

This allows native processes to get Windows-default error handling behavior (such as invoking the registered JIT debugger).

jeremyd2019 force-pushed the msys2-3_1_7-spawn-default-errormode branch from fdf1a51 to 496311d Compare December 10, 2020 18:33

jeremyd2019 closed this Dec 10, 2020

dscho mentioned this pull request Mar 31, 2021

msys2-runtime: upgrade to v3.2.0 msys2/MSYS2-packages#2414

Merged

dscho mentioned this pull request Mar 31, 2021

Turn off default Windows error reporting by default #35

Merged

jeremyd2019 deleted the msys2-3_1_7-spawn-default-errormode branch July 22, 2021 18:48

git-for-windows-ci mentioned this pull request Sep 6, 2022

[New msys2-runtime version] 19 new items git-for-windows/git#4009

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn #18

cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn #18

jeremyd2019 commented Nov 8, 2020 •

edited

Loading

dscho commented Nov 8, 2020

jeremyd2019 commented Nov 8, 2020 •

edited

Loading

jeremyd2019 commented Nov 9, 2020

jeremyd2019 commented Nov 9, 2020 •

edited

Loading

dscho commented Nov 10, 2020

dscho commented Nov 10, 2020

jeremyd2019 commented Nov 10, 2020

mingwandroid commented Nov 11, 2020

jeremyd2019 commented Nov 11, 2020

dscho commented Dec 3, 2020

jeremyd2019 commented Dec 3, 2020

jeremyd2019 commented Dec 10, 2020

dscho commented Mar 31, 2021

jeremyd2019 commented Mar 31, 2021

dscho commented Mar 31, 2021

cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn #18

cygwin: use CREATE_DEFAULT_ERROR_MODE in spawn #18

Conversation

jeremyd2019 commented Nov 8, 2020 • edited Loading

dscho commented Nov 8, 2020

jeremyd2019 commented Nov 8, 2020 • edited Loading

jeremyd2019 commented Nov 9, 2020

jeremyd2019 commented Nov 9, 2020 • edited Loading

dscho commented Nov 10, 2020

dscho commented Nov 10, 2020

jeremyd2019 commented Nov 10, 2020

mingwandroid commented Nov 11, 2020

jeremyd2019 commented Nov 11, 2020

dscho commented Dec 3, 2020

jeremyd2019 commented Dec 3, 2020

jeremyd2019 commented Dec 10, 2020

dscho commented Mar 31, 2021

jeremyd2019 commented Mar 31, 2021

dscho commented Mar 31, 2021

jeremyd2019 commented Nov 8, 2020 •

edited

Loading

jeremyd2019 commented Nov 8, 2020 •

edited

Loading

jeremyd2019 commented Nov 9, 2020 •

edited

Loading