Re: [lmi] PATCH: Switch to using std::filesystem

[Vadim Zeitlin]

On Tue, 11 May 2021 19:55:15 +0000 Greg Chicares <gchicares@sbcglobal.net> 
wrote:

GC> On 5/10/21 2:28 PM, Vadim Zeitlin wrote:
[...]
GC> >  My short-term TODO list is then:
GC> > 
GC> > 0. Fix the CI builds (I feel acutely uncomfortable without the safety they
GC> >    provide, so for me this is the highest priority).
GC> 
GC> Agreed. All automated tests must always succeed.

 Just to confirm: this is done now. I still got one transient error
recently due to a network error downloading Debian packages. If this is
going to happen again, I might do something about it (e.g. sleep for a bit
and retry if it happens again), but for now it doesn't happen often enough
(this was either the first or second time it ever did) to really be a
problem.

 BTW, as a part of doing this I've created a Docker containing with
everything needed for building lmi -- and nothing else. I could publish it
if it could be useful, please let me know if you think it could. As I think
I've already mentioned before, Docker is the modern replacement of chroot,
and while I still appreciate chroot simplicity for simple tests, Docker
does have several advantages, notably you always start from the same fresh
slate when using it, which was important for my tests.

GC> > 1. Use operator<=>() for fs::path -- and maybe other classes? Do you think
GC> >    I should restrict my patch to fs::path only or also use it for the 
other
GC> >    classes, such as:
GC> 
GC> It seems best to do them all at once.

 Will do.

GC> > 2. UTF-8 support for paths.
GC> 
GC> Agreed. And, BTW--I can't remember if I asked--what do you think
GC> of allowing UTF-8 in source files, for comments at least? I suppose
GC> that would work today if we just removed the 'test_coding_rules'
GC> code that forbids it.

 Yes, it really should work everywhere nowadays and I don't think we even
need to restrict them to the comments only, I'd like to use them in the
strings too, at the very least. As I've recently mentioned in another
reply, gcc still doesn't support using them in identifiers anyhow, so we're
not going to be able to do much more anyhow.

[file name validation]
GC> Agreed. Let's forget about doing that.

 I guess you'd prefer to check, and remove if necessary, any tests and
comments referring to it yourself? There is, of course, nothing urgent
about this.

GC> > GC> OTOH, testing the triple-dot example above gives
GC> > GC>   Warning: Unable to save 'Z:\opt\lmi.and another XYZ Corp 
Census.cns'.
GC> > GC>   [census_document.cpp : 94]
GC> > GC> and that's probably something lmi should trap.
GC> > 
GC> >  No, I think it's just a bug. But the problem is that I couldn't reproduce
GC> > it here with my existing MSVC build, i.e. the file gets saved without any
GC> > errors. I'll retest with MinGW binary slightly later, but I think this
GC> > might be actually something Wine-specific rather than a compiler-dependent
GC> > problem. And if this is really true, the impetus for doing anything here
GC> > becomes even smaller and, IMO, non-existent.
GC> 
GC> Agreed.

 I just wanted to confirm that I've tested the official binary built with
MinGW-w64 using makefiles and it works in the same way as the MSW one, i.e.
it allows saving the files with triple dots without any problems and so
what you've seen was definitely a Wine bug.

 Regards,
VZ

pgpG9B8LWsx6N.pgp
Description: PGP signature