Re: Sphinx parallel build error: UnicodeEncodeError: 'latin-1' codec can't encode characters in position 18-20: ordinal not in range(256)

[Mauro Carvalho Chehab <mchehab@kernel.org>]

Em Thu, 6 May 2021 19:04:44 +0200
Markus Heiser <markus.heiser@darmarit.de> escreveu:

> Am 06.05.21 um 18:46 schrieb Mauro Carvalho Chehab:
> > Em Thu, 6 May 2021 17:57:15 +0200
> > Markus Heiser <markus.heiser@darmarit.de> escreveu:
> >   
> >> Am 06.05.21 um 12:39 schrieb Michal Suchánek:  
> >>> When building HTML documentation I get this output:  
> >> ...  
> >>> [  412s] UnicodeEncodeError: 'latin-1' codec can't encode characters in position 18-20: ordinal not in range(256)  
> >> ...  
> >>>
> >>> It does not say which input file contains the offending character so I can't tell which file is broken.
> >>>
> >>> Any idea how to debug?  
> >>
> >> I guess the build host is a very simple container, what does
> >>
> >>     echo $LC_ALL
> >>     echo $LANG
> >>
> >> prompt?  If it is latin, change it to something using utf-8 (I recommend
> >> 'en_US.utf8').
> >>
> >> A UnicodeEncodeError can occour everywhere where characters are
> >> encoded from (internal) unicode to the encoding of the stream.
> >>
> >> By example:
> >>
> >> A print or log statement which streams to stdout needs to encode
> >> from unicode to stdout's encoding.  If there is one unicode symbol
> >> which can not encoded to stream's encoding a UnicodeEncodeError
> >> is raised.  
> > 
> > Hi Markus,
> > 
> > It shouldn't matter the builder's locale when building the Kernel
> > documentation (or any other documents built from other git trees
> > on other open source projects), as the Kernel's *.rpm document charset
> > won't change, no matter on what part of the globe it was built.
> > 
> > I vaguely remember about a change we made a couple of years ago
> > in order to address this issue.  
> 
> Hi Mauro :)
> 
> sure? .. what if the logger wants to log some symbols from the
> chines translated parts to stdout and the encoding of stdout is
> latin?
>
> In python the logger will raise a UnicodeEncodeError, this is
> what I know .. but I'm often wrong ;)

Yeah, Python (and almost all python apps) has a mad behavior when
it finds an unexpected character: instead of ignoring it, it
just crashes. On Sphinx, this is is even worse, as it blames
the parallel building, instead of pinpointing the real culprit.

In this specific case, crashing due to an invalid char sent to
the logger sounds pretty stupid to me, as what really matter
is the built documents.

> I remember we had some patches to the chinese translation
> these days, may be there is an issue the logger wants to report.
> 
>     Anyway I would always recommend to use utf-8.

Well, IMO, for things like logger, python or Sphinx should internally
be ding something similar to:

	$ iconv -t latin-1//IGNORE

e. g. if the system's charset is latin-1, it should ignore all charset
errors at the logger, converting the charset the best way it can
without crashing.

As, unfortunately, this is not happening, conf.py should do something
like hardcoding env{LANG}=<lang>.utf-8 (or something similar),
in order to to eliminate the risk of crashes.

Thanks,
Mauro