I'll try to clarify :)

I usually upgrade AS perl builds without rebuilding the packages that don't come with it. That's worked reliably since 5.8.0. However, when I first installed 5.8.7/AS813, the installed failed right towards the end. It didn't tell me what failed, but I am guessing that it was generating the HTMLised POD. It insisted on backing the install off, but left the previous 5.8.6 build broken to the point where perl simply wouldn't start. AS support suggested that I should always uninstall the previous build, (leaving the non-AS packages in place), before installng a new AS build. I attempted this with AS813 but it still failed at the end in the same way.

I reinstalled 5.8.6 over the top of what was left and things seemed okay. A few days later AS815 became available so I got that, uninstalled 5.8.6, again leaving the non distribution packages in place and attempted the AS815 install. It went through cleanly.

However, there were various instances of previously working code doing strange things--mostly resulting in a "free to wrong pool". I've seen this message a lot in the past with code that used threads badly, but these were simple non-threaded programs. The latest of which is the subject of the OP. This a highly cut-down version of a program that has worked fine since 5.8.3 and through several upgrades of GD, that suddenly started working one time but not the next. As you can see from the OP and zentara's post above, it is really very obscure when you can call a sub successfully once, or even twice to do exactly the same processing, from different places in the code--but not twice from the same place!

So, given the immediate preceding history of failed installed and broken restorations, I was unable in good conscience to 'point the finger' at 5.8.7. Hence my need to re-build everything clean to be sure.

My fear is, that as I have now built all those packages in conjunction with 5.8.7, whatever malady is the root cause of the segfault may now be compiled into, (via errors in the Perl header files?), all the (XS components of) packages that I built in conjunction with it.

It may be that the problem was unrelated to the header files in any way, but with perl 5.8.7 delta saying

There has been a fair amount of refactoring of the C source code,

can I be sure?

My fear is, that as I have now built all those packages in conjunction with 5.8.7, whatever malady is the root cause of the segfault may now be compiled into, (via errors in the Perl header files?), all the (XS components of) packages that I built in conjunction with it.

I think that fear is misplaced - I think it would be a rare case that a perl extension will need to be re-compiled (or benefit in any way from re-compiling) when you switch to perl 5.8.8.

Anyway, time will tell .... :-)

The error can still arise in perl 5.8.8. If you have a compiler, and Inline::C installed, and you run the Object Oriented Inline example from Inline's C-Cookbook.pod, you should get the "Free to wrong pool ..." error on Windows. That error still persists with perl 5.8.8 - but that's because it's the code in the Cookbook example that's in need of correction.

Cheers,
Rob

5.8.8 certainly cures the problem with my cut down code above, but not with the code from which it is derived :(

Now to go through the process of reducing it all over again.

Update:

If you get a moment, could you try wrapping the sub call in the OP code in a loop and calling say 10 times under 5.8.8? From what I am seeing, it appears that 5.8.8 allows me to call it twice without failure, but goes belly up after 4 or 5 times.

The cuprit appears to be something in GD, but a quick scan of GD.xs shows none of the bad things demo'd in in the OO Inline C example. It use Safemalloc/Safefree exclusively and make no calls to str(n)dup(), or any of the other dodgey crt string functions.

I tried upgrading GD to 2.30. It builds clean(ish), but segfaults when running it's testsuite. (And yes. I made sure I had the latest version of gdlib: v2.0.33 :)

---

Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.

Lingua non convalesco, consenesco et abolesco. -- Rule 1 has a caveat! -- Who broke the cabal?

"Science is about questioning the status quo. Questioning authority".

In the absence of evidence, opinion is indistinguishable from prejudice.