Re: My Emacs unicode 2 crash again when I do some *Replace String (M-%)*, I give the debug informations under gdb in the attachments.

[David Kastrup]

"Stephen J. Turnbull" <address@hidden> writes:

> Eli Zaretskii writes:
>
>  > The problem is, again, that the list of -fxxx options that, if
>  > disabled, will make the program much easier to debug, is not known
>
> We already have one (-fno-cross-jumps or similar).  Others can be
> added as people find out about them.  I'm simply saying that this
> information should be available to *all* Emacs developers (eg, by
> configuring with --marquis-de-sade-optimizations=no-thank-you), not
> just to those who happen to read every post David makes.

I've written this info into the file etc/DEBUG, too, but it would be
overoptimistic to assume that everybody reads that before starting
out.

I actually know no option with similarly devastating effects on
debugging productivity: if registers and values and stuff get
optimized away, that is basically a nuisance, but gdb tells you what
happens (at least with newer debug info formats, which value is where
is traced rather well).  I just fine-combed the other gcc options and
did not detect any option which would have the effect of leading the
debugging person to a completely wrong place: other stuff may make
debugging more adventurous, but as far as I can see, it will not cause
plausible lies from the debugger.

The crossjumps option ruins post-mortem debugging by very convincingly
making the backtrace point to the wrong failed assertion.  It does not
help to put a breakpoint on "abort" either.

It cost me several days of debugging once.

>  > So I don't think we know what to put in aclocal.

Well, not using -fno-crossjumping cost me several days of debugging,
it made me put the info in etc/DEBUG, and it cost another person a few
days before I again volunteered this information.

How much more wasted man-hours do you want, Eli, before "knowing" it
might be a good idea to use this by default?

Debugging failed assertions is not exactly a rare occurence.

I'd say what is good enough for etc/DEBUG is good enough for aclocal.
Of course, we'll need an autoconf check that this option is actually
supported by the gcc version in question.

-- 
David Kastrup