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

[Greg Chicares]

On 4/26/21 9:53 PM, Vadim Zeitlin wrote:
[...]
>  Sorry again for the inordinate delay with submitting this patch

Now that I've studied the first few commits, the real surprise is
that you and Ilya finished before the end of this decade. The
filesystem part is almost simple; coping with the still-evolving
utf-8-ization of C++ is a Herculean labor.

I had naively guessed that this was like the so-called Y2K problem,
which had an obvious solution, "just use YYYY everywhere", that was
simple and easy to implement. This problem likewise has an obvious
solution, "just use UTF-8 everywhere", which even has its own website:
  http://utf8everywhere.org/
but implementing it is complicated indeed. And we could do
  long int i = 1970
even at the dawn of the *nix epoch, as many had the foresight to
do even then; but if we write
  char const* c = "£";
then what we get is implementation-defined even today IIUC.

> I hope that at least the final version
> is now good enough to be applied.

It seems to be passing all tests except for that
  test failed:   '...-20210429T152844Z.' == '....'
problem mentioned yesterday, which we guessed was due to wine-5.x,
but now I seem to be seeing it with gcc-10 and
  $wine --version
  wine-4.0.3 (Debian 4.0.3-1)
Okay...I reenable my brutal workaround from yesterday (i.e.,
blocking only one pair of statements in one unit test), and
every known test passes, so I'll push it now and spend a few
more hours studying the details (and updating our proprietary
product repository) later.

The last commit pushed provides evidence suggesting that the
performance gap between 32- and 64-bit msw has nearly vanished.
I regard this with as much belief as I had in those experiments
that provided evidence that neutrinos travel faster than light;
but we are scientists, so we publish our measurements even if
we expect them to be irreproducible.