Seems like it works.  Thanks!

Try:

http://cvs.horde.org/diff.php?r1=1.9&r2=1.10&f=imp%2Flib%2FSpellChecker%2Faspell.php



?

I don't actually know any of the PHP developers, so I've tacked a 
comment onto one of the bugs I perceived to be applicable (and that 
was still allowing comments).  If one of you has an in with the folks 
at php.net, could you pursue this?



In the meantime, what about replacing...



         if (($ecode = proc_close($process)) != 0) {



with something like..



         if (strlen($out) == 0) {



or



         if (empty($out)) {



if there's a problem invoking aspell, $out stays empty, but $err gets 
filled.. so this seemed like it might work ok?

Well sure, that's what we're trying to do. What would you suggest? 
Maybe ask Ilia what people who _do_ use Oracle should do.

I can't figure out the difference between our test and development 
environments re: this issue.



As mentioned in...

http://bugs.php.net/bug.php?id=29123



     [16 Jul 2004 1:32am UTC] iliaa@php.net



     When sigchild is used the return status from pclose and

     proc_close() cannot be retrieved.



And --enable-sigchild is strongly recommended for PHP users using 
Oracle.  So.. if oracle is going to remain supported, I really think 
you have to find another way to judge whether aspell is working than 
checking the exit code provided by proc_close()

The code just makes it return 255 instead of -1 on the effected 
machine, which I've read might just be another way to say "-1".



I'd like to know why it works (returns exit code 0) in our dev 
environment but not our test environment.. but I haven't figured that 
out yet.  In the meantime, if I rebuild php without --enable-sigchild, 
the problem seems to go away.  That can't be the answer, tho.

Can you test if the:



$ret = (proc_close($proc) >> 8) & 0xff;



Trick works, and if it still works if you create a genuine failure? We 
can't just check for -1, since 0 _is_ being returned in other working 
PHP installs.

per http://marc.info/?l=imp&m=120853832309748&w=2



I think I've tracked the issue down.  IMP is looking for aspell to 
exit with a code of "0" to decide whether the spellcheck works.



imp/lib/SpellChecker/aspell.php ~ ln 67:



         if (($ecode = proc_close($process)) != 0) {

             require_once 'PEAR.php';

             if ($this->_encoding) {

                 $err = String::convertCharset($err, $this->_encoding, 
$charset);

             }

             return PEAR::raiseError('spellcheck failed: ' . $err, $ecode);

         }



proc_close() doesn't return results you think it's returning.  See...

   http://us2.php.net/manual/en/function.proc-close.php

   http://bugs.php.net/bug.php?id=17538



Apparently, it's especially faulty if you compiled php with 
--enable-sigchild, which you're told to do if you're trying to use 
oracle on linux.



The debugging stubs I placed in aspell.php clearly show that aspell is 
being invoked, and is returning the expected output.  If I change..



    proc_close($process)) != 0



to look for -1, spellchecking works just dandy.