Re: [PATCH v2] libctf: ctf_member_next needs to return (ssize_t)-1 on error

[Torbjorn SVENSSON <torbjorn.svensson@foss.st.com>]

On 2023-09-07 14:10, Nick Alcock wrote:
> On 30 Aug 2023, Alan Modra via Binutils outgrape:
> 
>> On Wed, Aug 30, 2023 at 10:34:05AM +0200, Torbjorn SVENSSON wrote:
>>> @Alan, any additional comments on this updated patch (except the missing
>>> semicolon that I mention below)?
>>
>> I'm leaving it to Nick Alcock to decide what to do here.
> 
> I agree that we should indeed be returning -1 from all functions that
> return an int (it used to, but ctf_id_t has to be an unsigned long). But
> I think it might be less disruptive to do so via a new
> ctf_set_errno_int() which is just like ctf_set_errno but returns an int
> rather than an unsigned long. That eliminates a lot of {}ery and makes
> each individual hunk smaller.

Ok, I can do that instead if that's considered the proper way.

> My concern is that it's really hard to validate all this -- can anyone
> think of a trick that would emit *consistent* warnings if we called
> return (ctf_set_errno()) from a function returning int? I mean this
> returning-int thing is a change I made ages ago, and despite making it
> *and* attempting to validate on 64-bit Windows I have clearly not got it
> right because it's drifted right out of correctness again.
>
> (Similarly, does anyone have a build/target triplet on which this goes
> wrong? because it's not going wrong on any of my mingw64 or cygwin tests
> as far as I can tell.)

I discovered the issue using the GCC12 package for arm-none-eabi that 
Arm released 
(https://developer.arm.com/downloads/-/arm-gnu-toolchain-downloads), but 
built using the x86_64-w64-mingw GCC compiler on Linux.

I've opened a ticket for the issue where I've attached 2 object files 
that you can use to reproduce the issue without needing to rebuild GCC + 
multlibs to verify the problem.
https://sourceware.org/bugzilla/show_bug.cgi?id=30836

> I kinda wish we could rely on having C11 -- type-generic macros are made
> for cases like this :(

Looks like it would be a nice fix indeed, but is there anything else 
that could be done to improve the situation without needing to go to C11?

Kind regards,
Torbjörn