[libpas] Report more actionable crash in pas_enumerator

• Component: New Bugs
• Status: RESOLVED FIXED

[Yusuke Suzuki]

[libpas] Report more actionable crash in pas_enumerator

[Yusuke Suzuki]

Created attachment 454060 [details]
Patch

[Mark Lam]

Comment on attachment 454060 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=454060&action=review

> Source/bmalloc/libpas/src/libpas/pas_utils.c:60
> +    pas_panic("%s:%d: %s: assertion %s failed.\n", filename, line, function, expression);

It looks like pas_panic will trample on the incoming args:
```
void pas_panic(const char* format, ...)
{
    static const bool fast_panic = false;
    if (!fast_panic) {
       ... 
        pas_vlog(format, arg_list);
    }
    __builtin_trap();
}
```

Can you also change pas_panic to not call __builtin_trap() directly, but call a NEVER_INLINE pas_trap(const char* filename, int line, const char* function, const char* expression) (similar to WTFCrashWithInfoImpl()).  This way we can at least see the line number in crash logs as well.  I think this will be useful.

[Mark Lam]

Comment on attachment 454060 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=454060&action=review

>> Source/bmalloc/libpas/src/libpas/pas_utils.c:60
>> +    pas_panic("%s:%d: %s: assertion %s failed.\n", filename, line, function, expression);
> 
> It looks like pas_panic will trample on the incoming args:
> ```
> void pas_panic(const char* format, ...)
> {
>     static const bool fast_panic = false;
>     if (!fast_panic) {
>        ... 
>         pas_vlog(format, arg_list);
>     }
>     __builtin_trap();
> }
> ```
> 
> Can you also change pas_panic to not call __builtin_trap() directly, but call a NEVER_INLINE pas_trap(const char* filename, int line, const char* function, const char* expression) (similar to WTFCrashWithInfoImpl()).  This way we can at least see the line number in crash logs as well.  I think this will be useful.

Never mind.  Yusuke reminded me offline that the caller of pas_assertion_failed_no_inline will be on the stack trace in the crash log.  So, the needed information is captured after all.

[EWS]

Committed r291007 (248183@main): <https://commits.webkit.org/248183@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 454060 [details].

[Radar WebKit Bug Importer]

<rdar://problem/89984164>

[Yusuke Suzuki]

Committed r291199 (248354@trunk): <https://commits.webkit.org/248354@trunk>