Ticket #1230 (closed bug: fixed)

Opened 7 years ago

Last modified 5 years ago

src/pmc/eval.pmc: Walk the fixups, locate globals and nullify the Sub PMC

Reported by: jkeenan Owned by: plobsing
Priority: normal Milestone:
Component: core Version: 1.7.0
Severity: medium Keywords:
Cc: Language:
Patch status: Platform:


This ticket moves into the Trac system an issue previously discussed in the RT system as  RT #46683.

139 =item C<opcode_t *invoke(void *next)>
141 Invokes the first subroutine in the eval code.
143 =cut
145 */
147     VTABLE opcode_t *invoke(void *next) {
148         PMC *sub = SELF.get_pmc_keyed_int(0);
149         return VTABLE_invoke(INTERP, sub, next);
150     }
152     VTABLE void destroy() {
153         /*
154          * If the compiled code contained any .sub (or .pcc.sub)
155          * subroutines, these subs got installed in the globals
156          * during compiling this bytecode segment.
157          *
158          * These globals still exist, calling them will segfault
159          * as the segment is destroyed now.
160          *
161          * RT#46683 walk the fixups for this segment, locate globals
162          *      and nullify the Sub PMC.
163          * This probably needs a pointer into the globals.
164          *
165          * OTOH - if the global exists - this eval pmc ought
166          *        to be alive and destroy isn't called.
167          */
168         PackFile_Segment  *seg;
169         PackFile_ByteCode *cur_cs;
170         Parrot_Sub_attributes *sub_data;


Change History

Changed 7 years ago by jkeenan

  • component changed from none to core

Changed 6 years ago by jkeenan

Is there anyone who could evaluate the issues raised in this ticket which originated in the RT system?

Thank you very much.


Changed 6 years ago by plobsing

  • owner set to plobsing

Changed 5 years ago by plobsing

  • status changed from new to closed
  • resolution set to fixed

If the sub still exists as a global, the code backing the sub should remain live. Invalidating the sub to avoid a segfault is very, very wrong.

The current solution is to have subs mark the eval_pmc backing them. This is the correct solution, as it exposes the object dependancy to the GC.

Note: See TracTickets for help on using tickets.