Re: [RFC/PATCH v2 1/1] cygwin: Add fast_lstat() and fast_fstat() functions

[Junio C Hamano <gitster@pobox.com>]

Mark Levedahl <mlevedahl@gmail.com> writes:

> Cygwin 1.7 is very different than the earlier, no longer supported,
> and no longer available Cygwin variants in many ways, but stat is one
> of them. Cygwin 1.7 uses Windows ACLs to represent file permissions,
> and therefore gets the file permissions directly from the underlying
> OS calls. Earlier Cygwin versions (attempted to) overlay POSIX
> permissions on Windows systems using extended attributes and other
> means, and in many cases had to resort to opening the file and
> examining it to determine executability. This is not true in 1.7.
>
> Therefore, your later patch would be expected to have much less
> benefit for 1.7 than for 1.5 (I don't detect *any* benefit on 1.7 when
> I set core.filemode=false). There are many choices, three are:
>
> a) Remove the win32 stat funcs, eliminating all of the troublesome
> code paths and maintenance burden (your original patch).
> b) Add your latest patch, with attendant complexity and maintenance
> burden, to support a version of Cygwin that is no longer available and
> was last updated over four years ago.
> c) Like b, except make this triggered only by a "CYGWIN_15" macro,
> limiting this to use by the legacy cygwin platform.

Let's do (a) in a single patch, then.

People who do want to keep running older Cygwin installation they
already have can revert the removal and rebuild Git, but the number
of people who have to do so will become only smaller over time if
older Cygwin versions are no longer available.

I presume that we _could_ add a CYGWIN_15 macro that conditionally
keeps the win32 lstat implementation and get_st_mode_bits() part,
and that might make it easier for folks with older Cygwin
installations, but I am not sure if it is worth it.

> I strongly vote for a, could support c, but fear b is just going to
> keep us chasing down bugs. Especially so when we consider that this
> patch can only speed things up when core.filemode=false, which mode:
> a) causes git to fail its test suite.
> b) breaks compatibility with Linux
> c) violates the primary goal of the Cygwin project, which is to
> provide a Linux environment on Windows.