bug#8254: race condition in dired.c's scmp function

[Paul Eggert <eggert@cs.ucla.edu>]

On 03/15/2011 04:36 AM, Juanma Barranquero wrote:

> there's adding complexity when it is not needed.

The patch subtracts complexity in one place (by removing global
variables) and adds it in another (by creating static inline
functions).  Whether the overall effect is to decrease complexity,
or to increase it, is debatable.  Either way, it's not much of
a change in complexity.

There are efforts underway to make Emacs multithreaded.  If that
happens, a change like this will be needed, as the existing
code is obviously not thread-safe.  I don't see any real downside
to installing this change in the trunk now.

> A (perhaps stupid) idea: would it be possible to define
> -DENABLE-CHECKING alternate versions of DOWNCASE and UPCASE1 which do
> some additional checking for side effects?

I plan to implement that sort of suggestion, but in a different
way, by adding an --enable-gcc-warnings option to 'configure',
which will cause it to pass extra options to GCC to catch
this sort of problem.

This option is already in used in several other projects, such
as GNU coreutils, and Emacs would benefit from it as well.
The option will be disabled by default, though, so that the warnings
don't surprise people who don't expect them.