Re: [PATCH] Add Windows support for backtrace_functions (MSVC only)

On 10/30/2025 3:52 AM, Jakub Wartak wrote:
> On Thu, Oct 30, 2025 at 10:40 AM Bryan Green <dbryan.green@gmail.com> wrote:
>>
>> On 10/30/2025 3:37 AM, Álvaro Herrera wrote:
>>> On 2025-Oct-30, Jakub Wartak wrote:
>>>
>>>> Hi Bryan, cfbot is red. I'm was fan of having those tests for this
>>>> (bring complexity and we didn't have tests for Linux backtrace
>>>> anyway), but now MINGW win32 is failing on those tests where the
>>>> feature is not present:
>>>
>>> I hate to say this after the code is written, but I think we should not
>>> put any tests in the first step.  I predict that these are going to be
>>> enormously brittle and that we'll waste a lot of time making them
>>> stable.  I think we should commit the Windows support for backtraces
>>> first, then consider whether we actually want TAP tests for the overall
>>> feature.  We've gone several years with glibc backtrace support without
>>> any tests -- why do we think the Windows implementation thereof _must_
>>> necessarily have them?
>>>
>> It will not bother me to remove them.  It was my first effort at writing
>> TAP tests, so it was a nice learning experience.
> 
> Well, that was a typo on my part (stupid me), I wanted to write: I was
> NOT a fan of having those tests for this (in first place) - sorry for
> confusion!
> 
> Anyway we have test because I think Michael and Euler triggered this
> but earlier i've tried to persuade NOT to do this (see: `Also is it
> worth it to test that setting backtrace_funciton=FOO really emits
> .*FOO.* in log message cross-platform way?`), anyway Bryan implemented
> this and it looks like v3 has just turned [gG]reen ;)
> (https://cirrus-ci.com/build/6001832838823936)
> 
> -J.
The tests are easy enough to get rid of. I think Alvaro has a good idea 
of committing the windows support for backtraces and then consider 
whether we want TAP tests or not.  I will make a v4 patch without the 
TAP tests unless someone strongly objects.

The tests don't really test whether this code would be the cause of 
cause of a problem, they mainly test whether you are getting the correct 
output in your backtrace.  If you have a pdb file, you should get 
filenames and linenumbers in addition to addresses and symbols.  If you 
don't have a pdb file you will only get export function symbols and 
addresses.  So, even if you should have a pdb and don't...you still get 
something useful.

Bryan Green