Add runtime config option to change exception trapping behavior #1538

Marenz · 2016-04-07T13:40:40Z

This way you can do

extern extern (C) __gshared bool rt_trapExceptions;
static this ( )
{
        rt_trapExceptions = false;
}

to change the runtime behavior before main() is called.

CyberShadow · 2016-04-07T19:48:38Z

Interesting, I thought rt_trapExceptions applies to unit tests / constructors as well. I think it should.

Perhaps update its documentation to reflect that it is now also settable from a static constructor?

Marenz · 2016-04-07T20:03:01Z

Updated documentation for the variable

Marenz · 2016-04-08T19:55:05Z

Updated unit test to account for changed line numbers

Marenz · 2016-04-15T17:15:12Z

Ping.

Marenz · 2016-05-01T11:00:25Z

Ping

Marenz · 2016-05-01T11:03:58Z

It would be nice to get some feedback or at least any reaction to this.

rainers · 2016-05-03T09:12:40Z

I have not tried, but I don't think it works to set the flag in a ctor: there is an outer tryExec(&runAll) that will still catch excetions and errors (including those from static ctors).
I guess it might make sense together with applying the same flag to fibers, though.

CyberShadow · 2016-05-03T09:20:03Z

I have not tried, but I don't think it works to set the flag in a ctor: there is an outer tryExec(&runAll) that will still catch excetions and errors (including those from static ctors).

That's what I thought too, but it appears to be (no longer?) so.

Marenz · 2016-05-05T15:13:13Z

You were right, it didn't work. I did test it and something did work before but I guess not this.
I updated the PR so that it is a runtime config option now. I tested it with

void main()
{
    assert(1==2);
}

dub build
Performing "debug" build using dmd for x86.
test ~master: building configuration "application"...
Linking...

(gdb) r
Starting program: /home/marenz/projekte/test/test 
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/i386-linux-gnu/i686/cmov/libthread_db.so.1".
core.exception.AssertError@source/app.d(3): Assertion failure
----------------
??:? _d_assert [0x806654e]
??:? void app.__assert(int) [0x8066494]
source/app.d:3 _Dmain [0x806644c]
??:? _D2rt6dmain211_d_run_mainUiPPaPUAAaZiZ6runAllMFZ9__lambda1MFZv [0x80668aa]
??:? void rt.dmain2._d_run_main(int, char**, extern (C) int function(char[][])*).tryExec(scope void delegate()) [0x80667ff]
??:? void rt.dmain2._d_run_main(int, char**, extern (C) int function(char[][])*).runAll() [0x8066866]
??:? void rt.dmain2._d_run_main(int, char**, extern (C) int function(char[][])*).tryExec(scope void delegate()) [0x80667ff]
??:? _d_run_main [0x8066792]
??:? main [0x80664d3]
??:? __libc_start_main [0xb7d9ea62]
[Inferior 1 (process 28082) exited with code 01]
(gdb) bt
No stack.

(gdb) r --DRT-noTrapExceptions=1
Starting program: /home/marenz/projekte/test/test --DRT-noTrapExceptions=1
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/i386-linux-gnu/i686/cmov/libthread_db.so.1".
uncaught exception
dwarfeh(224) fatal error

Program received signal SIGABRT, Aborted.
0xb7fdcd40 in __kernel_vsyscall ()
(gdb) bt
#0  0xb7fdcd40 in __kernel_vsyscall ()
#1  0xb7db3367 in __GI_raise (sig=sig@entry=6) at ../nptl/sysdeps/unix/sysv/linux/raise.c:56
#2  0xb7db4a23 in __GI_abort () at abort.c:89
#3  0x08066a78 in _d_throwdwarf ()
#4  0x0806654f in _d_assert ()
#5  0x08066495 in app.__assert() ()
#6  0x0806644d in D main () at source/app.d:3
#7  0x080668ab in rt.dmain2._d_run_main() ()
#8  0x08066838 in rt.dmain2._d_run_main() ()
#9  0x08066867 in rt.dmain2._d_run_main() ()
#10 0x08066838 in rt.dmain2._d_run_main() ()
#11 0x08066793 in _d_run_main ()
#12 0x080664d4 in main ()

Or outside GDB:

➜  test  ./test --DRT-noTrapExceptions=1
uncaught exception
dwarfeh(224) fatal error
[1]    28317 abort (core dumped)  ./test --DRT-noTrapExceptions=1
➜  test  ./test                         
core.exception.AssertError@source/app.d(3): Assertion failure
----------------
??:? _d_assert [0x806654e]
??:? void app.__assert(int) [0x8066494]
source/app.d:3 _Dmain [0x806644c]
??:? _D2rt6dmain211_d_run_mainUiPPaPUAAaZiZ6runAllMFZ9__lambda1MFZv [0x80668aa]
??:? void rt.dmain2._d_run_main(int, char**, extern (C) int function(char[][])*).tryExec(scope void delegate()) [0x80667ff]
??:? void rt.dmain2._d_run_main(int, char**, extern (C) int function(char[][])*).runAll() [0x8066866]
??:? void rt.dmain2._d_run_main(int, char**, extern (C) int function(char[][])*).tryExec(scope void delegate()) [0x80667ff]
??:? _d_run_main [0x8066792]
??:? main [0x80664d3]
??:? __libc_start_main [0xb74a2a62]

Marenz · 2016-05-05T16:42:43Z

@MartinNowak trying a direct hilight

adamdruppe · 2016-05-05T16:47:20Z

Yes, yes, yes, merge this!

BTW if it isn't merged, since it is an extern(C) variable, you can also redefine it yourself and instruct the linker to use your version instead of the druntime one. hacky sure but hey.

extern(C) __gshared bool rt_trapExceptions = false;

compile:
dmd test.d -L-zmuldefs -g

but this is a trivial change that has clear value, looks very good to me.

dnadlinger · 2016-05-06T15:24:38Z

How do we test the command line switch?

Marenz · 2016-05-06T15:40:30Z

@klickverbot I wrote it above already: you pass --DRT-noTrapExceptions=1 to your application (one of the ways) and any assert failure will cause a coredump now instead of just outputting the stack trace.

dnadlinger · 2016-05-06T15:48:00Z

@Marenz: I realised that. What I was pointed out is that this PR is currently missing regression tests. Due to the nature of the feature, you'd probably need to add a new standalone test executable to test/.

Marenz · 2016-05-06T15:49:23Z

Do you want to test the switch itself or just whether the runtime behaves correctly when the option is enabled?

Marenz · 2016-05-06T15:49:51Z

I am asking because the parsing etc of that switch is all functionality that is already present.

dnadlinger · 2016-05-06T15:52:12Z

I want something that breaks if the code part of this PR is reverted (well, except for the line numbers^^).

Marenz · 2016-05-06T15:55:06Z

Will that test need to be cross-platform?

leandro-lucarella-sociomantic · 2016-05-09T13:16:15Z

src/rt/dmain2.d

@@ -477,6 +474,13 @@ extern (C) int _d_run_main(int argc, char **argv, MainFunc mainFunc)
            result = (result == EXIT_SUCCESS) ? EXIT_FAILURE : result;
    }

+    import rt.config : rt_configOption;
+
+    if (rt_configOption("noTrapExceptions") !is null)


I suggest using options without negative: trapExceptions

Gladly, but that means the default will be changed, right? Is that acceptable @MartinNowak

No, that's not necessary (nor desirable, I think). Just make it default to 1, not 0.

Certainly not desirable :) It would mean that programs will segfault instead of showing an exception stack trace on uncaught exceptions by default.

We have different definitions of "desirable", but sure, I see if I can change it

It looks like you're using the result of rt_configOption like a bool here... which means that --DRT-noTrapExceptions=1 and --DRT-noTrapExceptions=0 will both do the same thing. This is not ideal.

I looked to see if we have a precedent for this, and here is how #1798 added callStructDtorsDuringGC, also a boolean option:

druntime/src/rt/lifetime.d

Lines 49 to 58 in 639878d

extern (C) void lifetime_init()

{

// this is run before static ctors, so it is safe to modify immutables

import rt.config;

string s = rt_configOption("callStructDtorsDuringGC");

if (s != null)

cast() callStructDtorsDuringGC = s[0] == '1' || s[0] == 'y' || s[0] == 'Y';

else

cast() callStructDtorsDuringGC = true;

}

I suggest to parse this option in a similar way, which would also allow e.g. overriding the default behaviour of disabling rt_trapExceptions when running under a debugger on Windows. That also solves the "double negation" issue. If you're feeling up for it, refactoring the bool checks into a new function in the rt.config module, which also rejects invalid values, would be a bonus.

adamdruppe · 2017-06-20T12:36:55Z

why is this still in limbo?

CyberShadow · 2017-06-20T15:26:10Z

Looks good to me. @MartinNowak ?

CyberShadow

@Marenz Are you still up for this? Since apparently no one has any feedback, let's clean this up & merge it.

A changelog entry with an example of enabling this via the command-line or environment would be nice.

BTW, this is closely related to #1533 and #1673.

wilzbach · 2018-06-26T05:51:29Z

Superseded by #2035

leandro-lucarella-sociomantic · 2018-06-26T12:27:23Z

It is superseded even if the title of the new issue says is for the CLI but also via a similar trick in the source code, right?

wilzbach · 2018-06-26T13:54:39Z

AFAICT this stagnated/orphaned PR was about allowing --DRT-noTrapExceptions=1 (or environment config) and #2035 is exactly about this too (just that a different naming was used).
__gshared bool rt_trapExceptions already exists in DRuntime.

Marenz mentioned this pull request Apr 7, 2016

Make catching of exceptions from Fibers optional #1288

Closed

Marenz force-pushed the runttrap branch from 41296d4 to d250f4d Compare April 7, 2016 19:43

Marenz force-pushed the runttrap branch from d250f4d to 2adab78 Compare April 7, 2016 20:02

Marenz force-pushed the runttrap branch from 2adab78 to d1600bc Compare April 8, 2016 19:53

Marenz force-pushed the runttrap branch 2 times, most recently from e9bf524 to c7df340 Compare May 5, 2016 15:04

Add runtime config option to change exception trapping behavior

c7df340

Marenz changed the title ~~Allow changing of exc. trapping behavior from module c'tors~~ Add runtime config option to change exception trapping behavior May 5, 2016

leandro-lucarella-sociomantic reviewed May 9, 2016
View reviewed changes

Burgos mentioned this pull request May 9, 2017

Allow rt_trapExceptions to be set from command line #1674

Closed

nemanja-boric-sociomantic mentioned this pull request Jun 6, 2017

Disable rt_trapExceptions when started inside a debugger on Linux #1811

Closed

CyberShadow suggested changes Jul 15, 2017

View reviewed changes

CyberShadow mentioned this pull request Jul 15, 2017

fix Issue 15886 - Add an uncaught exception handler #1533

Closed

dlang-bot added Needs Rebase needs a `git rebase` performed stalled Needs Work labels Jan 1, 2018

rainers mentioned this pull request Jan 27, 2018

Fix Issue 18220 - Allow rt_trapexceptions to be set from the CLI #2035

Merged

wilzbach closed this Jun 26, 2018

	extern (C) void lifetime_init()
	{
	// this is run before static ctors, so it is safe to modify immutables
	import rt.config;
	string s = rt_configOption("callStructDtorsDuringGC");
	if (s != null)
	cast() callStructDtorsDuringGC = s[0] == '1' \|\| s[0] == 'y' \|\| s[0] == 'Y';
	else
	cast() callStructDtorsDuringGC = true;
	}

Uh oh!

Add runtime config option to change exception trapping behavior #1538

Add runtime config option to change exception trapping behavior #1538

Uh oh!

Conversation

Marenz commented Apr 7, 2016

Uh oh!

CyberShadow commented Apr 7, 2016

Uh oh!

Marenz commented Apr 7, 2016

Uh oh!

Marenz commented Apr 8, 2016

Uh oh!

Marenz commented Apr 15, 2016

Uh oh!

Marenz commented May 1, 2016

Uh oh!

Marenz commented May 1, 2016

Uh oh!

rainers commented May 3, 2016

Uh oh!

CyberShadow commented May 3, 2016

Uh oh!

Marenz commented May 5, 2016

Uh oh!

Marenz commented May 5, 2016

Uh oh!

adamdruppe commented May 5, 2016

Uh oh!

dnadlinger commented May 6, 2016

Uh oh!

Marenz commented May 6, 2016

Uh oh!

dnadlinger commented May 6, 2016

Uh oh!

Marenz commented May 6, 2016

Uh oh!

Marenz commented May 6, 2016

Uh oh!

dnadlinger commented May 6, 2016

Uh oh!

Marenz commented May 6, 2016

Uh oh!

leandro-lucarella-sociomantic May 9, 2016

Choose a reason for hiding this comment

Uh oh!

mathias-baumann-sociomantic May 10, 2016

Choose a reason for hiding this comment

Uh oh!

leandro-lucarella-sociomantic May 10, 2016

Choose a reason for hiding this comment

Uh oh!

CyberShadow May 10, 2016

Choose a reason for hiding this comment

Uh oh!

mathias-baumann-sociomantic May 10, 2016

Choose a reason for hiding this comment

Uh oh!

CyberShadow Jul 15, 2017

Choose a reason for hiding this comment

Uh oh!

adamdruppe commented Jun 20, 2017

Uh oh!

CyberShadow commented Jun 20, 2017

Uh oh!

CyberShadow left a comment

Choose a reason for hiding this comment

Uh oh!

wilzbach commented Jun 26, 2018

Uh oh!

leandro-lucarella-sociomantic commented Jun 26, 2018

Uh oh!

wilzbach commented Jun 26, 2018

Uh oh!

Uh oh!