[cfe-dev] Asserts in Clang crashes my Kdevelop

[Richard Smith]

On Mon, Jan 26, 2015 at 11:46 PM, Olivier J. G. <olivier.jg at gmail.com>
wrote:

> void Lexer::lexVerbatimBlockFirstLine(Token &T) {
>> again:
>>   assert(BufferPtr < CommentEnd);
>>
>> So, failing that assertion is equivalent to memory OOB access.
>>
>> Barring an enormous push to stabilize Clang and make it less crashy, you
>> have two options:
>> 1. Try to recover from assertion failures and segfaults with
>> CrashRecoveryContext
>> 2. Rig up a separate process to run clang in from kdevelop
>>
>> A separate process will be more stable but far more effort.
>>
>> You'd think compilers wouldn't crash very often, but take a look at the
>> list of crashers found by AFL:
>> http://llvm.org/bugs/buglist.cgi?quicksearch=fuzz&list_id=66187
>> While these are all bugs we'd like fixed, it's not clear how soon that
>> will happen. Until then, KDevelop and IDEs in general will need to cope
>> with compiler crashes somehow.
>>
>>
> While it seems like this assertion failure would lead to a crash, I should
> note that I haven't run into any crashes during AST traversal when clang is
> in release mode. OTOH, clang compiled in debug mode will fairly regularly
> come up with assertions during AST traversal.
>
> For the initial parsing we see plenty of crashes and asserts in both debug
> and release, but libclang already runs those in a CrashRecoveryContext, so
> it's not usually an issue (unless you're trying to run KDevelop in gdb...).
>
> In short, KDevelop/kdev-clang runs smoothly with release-build clang and
> is not really usable when clang is built w/ assertions.
>

Nonetheless, if you can provide us with testcases for the crashes, we'd
appreciate them.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20150127/f754f595/attachment.html>