Re: Odd build error in 1.85 pre-release - filesystem

From: Andrey Semashev (andrey.semashev_at_[hidden])
Date: 2024-02-23 14:06:15

On 2/23/24 16:53, Peter Dimov via Boost wrote:
> Andrey Semashev wrote:
>> On 2/23/24 05:07, Tom Kent via Boost wrote:
>>> When building for windows against the 1.85.0 nightly snapshot, I'm
>>> seeing an odd error with no info around it:
>>>
>>>
>>>     - type_erasure             : building
>>>     - url                      : building
>>>     - wave                     : building
>>>
>>> ...patience...
>>> ...patience...
>>> ...patience...
>>> ...patience...
>>> ...patience...
>>> ...patience...
>>> ...patience...
>>> [errno 22] failed to scan file '': Invalid argument...found 39744 targets...
>>>
>>>
>>> Upping the verbosity in B2, I was able to find that it appears to
>>> happen in
>>> filesystem:
> ...
>> It looks like the error happens in b2 while it is scanning headers for
>> dependencies. I don't know b2 internals enough to understand why that
>> happens and whether there is something specific in Boost.Filesystem to trigger
>> it. In any case, boost\filesystem\detail\header.hpp is not new and hasn't
>> changed for more than a year, so I don't think it's the header. I think, this
>> should be reported to Boost.Build.
>>
>> https://github.com/bfgroup/b2/issues
>>
>> PS: I'm not seeing this locally or in CI:
> 
> The problem is not in Filesystem, but in Wave. I've verified that
> 
> b2 --without-wave
> 
> does not give the error, and
> 
> b2 --with-wave
> 
> does.
> 
> For me, the latter outputs
> 
> ...patience...
> ...patience...
> [errno 22] failed to scan file '': Invalid argument[errno 2] failed to scan file 'boost/wave\': No such file or directory...found 6224 targets...

Indeed, and I believe this is triggered by these lines:

https://github.com/boostorg/wave/blob/737c1e07eedda7f0131ec601df1a34300a1f38f0/include/boost/wave/cpp_iteration_context.hpp#L101-L102

If I change those comments to "#include <header>" and "#include
"header"" then the error disappears.

The bug is in b2 though, as it should not be triggered by comments (or
anything that isn't an actual #include directive).