Re: [RFC 0/4] Make strsignal and strerror async-signal-safe

[Adhemerval Zanella <adhemerval.zanella@linaro.org>]

On 14/05/2020 12:11, Zack Weinberg wrote:
> On Thu, May 14, 2020 at 8:47 AM Adhemerval Zanella via Libc-alpha
> <libc-alpha@sourceware.org> wrote:
>> On 14/05/2020 09:05, Florian Weimer wrote:
>>> * Adhemerval Zanella via Libc-alpha:
>>>
>>>>   2. Avoid to try translate the returned message.
>>>
>>> No more translation is a significant change.  Is this really
>>> appropriate?
>>>
>>> Using strerror output in translated strings is very common:
>>>
>>> <https://codesearch.debian.net/search?q=_%5C%28.*strerror&literal=0>
>>
>> I take that once we have proper symbols that provides the translation
>> functionality, making them async-signal-safe is an welcome improvement.
>>
>> The glibc itself uses the __sys_* access on libSegFault due this
>> shortcoming.  And I am not if other programs are really aware of
>> the shortcoming of the always translatable error strings.
> 
> I think I'm with Florian here.  There is lots of existing code that
> expects strerror() to translate, and _most_ of the time it's not being
> called from an async signal handler, making everyone change their code
> to use strerror_l (and get a locale_t to use it with) just to get back
> the behavior they're used to seems like too much churn.

Alright, I took this semantic change would to make them async-signal-safe
allow to be more beneficial in the long term (in both code complexity and
more concise API) but looks like we should go to newer symbols instead.

> 
> It would be nice if there was a way for strerror to detect that it
> wasn't safe for it to call dcgettext or allocate memory.  In principle
> we ought to be able to try-lock the "I might need to initialize the
> global locale now" lock but, looking at the guts of dcgettext, there's
> several different locks involved and it's not clear to me that we
> could make this work without a major overhaul of libintl.
> 
> (Giant-hammer idea: have sigaction wrap all signal handlers with a
> routine that sets a thread-local flag while calling the signal
> handler.  siglongjmp and swapcontext would need to restore this flag.)

The sigaction wrap was a suggestion I proposed to Florian some time ago
as a way to detect a code was running in a signal handler.  FreeBSB does
this for the its libpthread analogous and its adds some complexity 
(it uses a static list and it requires some locking to handle concurrent
access), but I think it should be feasible.