[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: cygwin breakage
Re: cygwin breakage
Tue, 19 Oct 2004 00:24:51 -0700
On Mon, Oct 18, 2004 at 09:51:08AM +0100, Gary V. Vaughan wrote:
> >+ eval $show ""if test "x`/usr/bin/sed 1q .libs/libfoo2.exp`" = xEXPORTS;
> >then cp .libs/libfoo2.exp .libs/cygfoo2-0.dll.def; else echo EXPORTS >
> >.libs/cygfoo2-0.dll.def; cat .libs/libfoo2.exp >> .libs/cygfoo2-0.dll.def;
> >+ /usr/bin/sed 1q .libs/libfoo2.exp
> >+ echo if test xlibfoo2_LTX_foo2 = xEXPORTS
> >if test xlibfoo2_LTX_foo2 = xEXPORTS
> >eval: 1: Syntax error: "then" unexpected
> >The two pairs of quotes cancel, so the variable expansion evals
> >-eval \$show \"$func_quote_for_eval_result\"
> >+eval \$show "$func_quote_for_eval_result"
> >plugs the failure in this case, but I think it unfixes the problem Peter
> >trying to fix. Indeed, ``$show "$my_arg"'' is equivalent. Hmmm.
> I had come to the same conclusion, and after *a lot* of buggering about
> with quote
> substitution and sed in func_quote_for_eval managed to come up with the
> sed magic needed to protect the expression returned in
> $func_quote_for_eval_result from early expansion...
> As it turns out, your untested patch at:
> is a much better generalised analysis and cleaner patch that mine.
> As I had hoped, this also paves the road to cleaning up all the
> $show/$run cruft on HEAD. Excellent catch!
Thank you. I also wrote a test case that validates the behavior of
func_quote_for_eval with the invariant I proposed in that message, using a
number of metacharacters. I will clean it up for inclusion.
There remains the issue of expanding parameters in the ``$show'' of an ``eval
$run''ed command. func_quote_for_eval'ing the command prevents word splitting,
command substitution, and parameter expansion. We need something else, call it
func_quote_for_expand, that only stops word splitting and command substitution.
I have attached a sample implementation. With something like it, one writes:
eval "$show $func_quote_for_expand_result"
$run eval "$cmd"
> I've successfully run the testsuite on my powerbook, and the failing
> mdemo tests are now working on cygwin. To be doubly sure there are no
> regressions, I am running the entire testsuite on cygwin before I
> commit. I expect it to take another 12-18 hours for that to complete
> (assuming Windows 98 can stay up that long under heavy processor load ;-)).
The Autoconf test suite has taken up residence on my 486/33 laptop. It appears
on track for completion within sixteen hours.
Description: Bourne shell script