Re: Temporary files == slowness

[Eric Sunshine]

Hello, thanks for responding.

Paul Eggert wrote:
> I'm surprised the problem ever went away, as the subtle quoting
> differences between

It didn't.  I was misled by a change to the way that Autoconf picks a shell.

> Unless someone else pipes up with the real answer then I suppose we'll
> have to get to the bottom of this ourselves.

I have already done a good deal of additional debugging since posting 
the original email.  It turns out that the problem is that Autoconf CVS 
now chooses to use the older factory-supplied NextStep shell, rather 
than the newer, modern version of Bash which I have installed.  Autoconf 
versions up to 2.59 correctly choose the newer shell.

The reason that Autoconf CVS chooses the older shell is because it is 
only looking for a shell which implements shell functions (which the 
NextStep shell does), and the factory-supplied shell is the first one it 
finds.  The reason that 2.59 (and earlier) chooses the newer Bash is 
that it is looking for a shell which implements $LINENO (which the 
NextStep shell does _not_).  Therefore, 2.59 prefers Bash (since it 
implements $LINENO) over the factory-supplied NextStep shell.

I think the appropriate fix is to prefer a shell which implements both 
functions and $LINENO if such a shell can be found.  If not, then it 
should pick a shell which implements one or the other.  I am planning on 
submitting a patch which will implement this heuristic.  Not only will 
this make me happy (since running modern-day configure scripts is 
impossibly and painfully slow with the old NextStep shell), but it also 
just seems like the correct approach (i.e. preferring a more functional, 
modern shell over a less functional, slower one).

-- ES