Re: xsubpp and Tk ==> segfault

Ilya Zakharevich writes:
>
> I recall that the reason why void code may be marked as XSRETURN(1) is
> that XS TUTORIAL recommended to mark XSUB as void if you manually put

It is not perlxstut, it is perlxs.

> in stack manipulation! I do not know why this horrible practice is
> blessed, but we need a workaround which would allow this mess, and
> would not make an XSUB

[Add: ] ... segfault if some Perl code was called in between.

As I found out, the only difference between an XSUB with a CODE:
section which is declared `void' and one which is declared "SV *" is
that the first one does not contain a line:
	SV * RETVAL;
Is there any trouble if such a line is present?

If nobody will strongly object to it, I will remove an advice to
mark a CODE:-XSUBs as `void' from the POD(s).

I can also add a warning to xsubpp if a `void' XSUB sets ST(*). What
about it? (Warning will be issued only once per .xs .)

> Solution: Make the check for CODE section stricter: check for actuall
> occurence of "ST(" (with embedded spaces) before making the decision
> to XSRETURN(1).

The following patch is checked with Perl 3_17 and Tk (it checks for ST
at the LHS as an assignment):

Enjoy,

p5p-msgid: <199701080825.DAA15813@monk.mps.ohio-state.edu>

diff --git a/lib/ExtUtils/xsubpp b/lib/ExtUtils/xsubpp
index d43eb1c..76e45d6 100755 (executable)
--- a/lib/ExtUtils/xsubpp
+++ b/lib/ExtUtils/xsubpp
@@ -842,6 +842,7 @@ while (fetch_para()) {
 
     $PPCODE = grep(/^\s*PPCODE\s*:/, @line);
     $CODE = grep(/^\s*CODE\s*:/, @line);
+    $EXPLICIT_RETURN = $CODE && ("@line" =~ /\bST\s*\([^;]*=/ );
     $ALIAS  = grep(/^\s*ALIAS\s*:/,  @line);
 
     # print function header
@@ -1025,7 +1026,7 @@ EOF
 #      croak(errbuf);
 EOF
 
-    if ($ret_type ne "void" or $CODE) {
+    if ($ret_type ne "void" or $EXPLICIT_RETURN) {
         print Q<<EOF unless $PPCODE;
 #    XSRETURN(1);
 EOF