Re: intl: plural form evaluation crashes — worth fixing?

[Adhemerval Zanella Netto]

On 06/10/23 14:51, Bruno Haible wrote:
> Adhemerval Zanella Netto wrote:
>> Yes, I think both issues are worth fixing.
> 
> OK, I'll then enter them in bugzilla and prepare patches.
> Can you please give me edit access (not only creation and comment access) to
> bugzilla, per https://sourceware.org/glibc/wiki/Bugzilla%20Procedures § B0 ?
> 
>> In fact, building dcigettext.c
>> with -fstack-usage show that plural_eval uses about 48 bytes and with
>> EVAL_MAXDEPTH being 100 it means about worst case of ~4.8MB which is
>> still quite large.
> 
> No no. 48 bytes/frame * 100 frames = 4.8 KB, not 4.8 MB. Even threads on
> very small systems have 16 KB at least.

Oops, yeah I mean KB indeed.  But it is still somewhat large for a *single*
function call (recall that we don't when in the stackframe the thread will 
ending calling the libc).

> 
>> Do we really EVAL_MAXDEPTH set to 100?
> 
> musl libc, which is used in "smaller" systems than glibc, has set it to 100
> too.

Fair enough.