view etc/future-bug @ 58818:f8cddae7d959

* gtkutil.c: Include signal.h and syssignal.h. (xg_get_file_name): Block and unblock __SIGRTMIN if defined. * alloc.c: If HAVE_GTK_AND_PTHREAD, include pthread.h, new variables main_thread and alloc_mutex, define (UN)BLOCK_INPUT_ALLOC to use alloc_mutex to protect emacs_blocked_* calls and only do (UN)BLOCK_INPUT in the main thread. If not HAVE_GTK_AND_PTHREAD, (UN)BLOCK_INPUT_ALLOC is the same as (UN)BLOCK_INPUT. (emacs_blocked_free, emacs_blocked_malloc) (emacs_blocked_realloc): Use (UN)BLOCK_INPUT_ALLOC. (uninterrupt_malloc): Initialize main_thread and alloc_mutex. (reset_malloc_hooks): New function. * lisp.h: Declare reset_malloc_hooks. * emacs.c (Fdump_emacs): Call reset_malloc_hooks. * keyboard.c: Conditionally include pthread.h (handle_async_inpu, input_available_signalt): If not in the main thread, block signal, send signal to main thread and return.

Date: Tue, 23 Feb 2199 21:03:50 -0600
From: Karl Fogel <[e-mail]>
To: [e-mail]
Subject: M-x search-backward-in-time broken...
X-Windows: you'll envy the dead.

In GNU Emacs 51.70.4 (i9986-unknown-linux-gnu, X toolkit) of Sat Feb  20 2199 on floss
configured using `configure  --with-x-toolkit=yes'

        The `search-backward-in-time' function appears to be broken in
Emacs 51.70.

        Unfortunately, I can never seem to start the debugger early
enough to catch the error as it happens.  However I have traced the
problem through source by eye, and it looks like `time-forward' can't
handle negative arguments anymore.  This is consistent with other
symptoms: for example, `undo' (which since 51.25 has worked by passing
a negative arg to `time-forward') is also broken.  However, `do' still
works -- it seems that `time-forward' continues to handle positive
arguments just fine.

        No one here-and-now can figure out how to fix the problem,
because the code for `time-forward' is so hairy.  We're using M-x
report-future-emacs-bug to request that you folks include more
comments when you write it (sometime in 2198 as I recall).

        Thanks!

-Karl Fogel   <[e-mail]>


P.S. You'll be pleased to know that since (time-forward N) still works
     for N >= 0, we've used it to pre-emptively update configure.in.
     Emacs now configures and builds on every platform that will ever
     be made.  It wasn't easy, but at least that's one problem out of
     the way for good.  If you'd like the patch, just ask.