Debugging B::C, the workaround (part 3)

In the first B::C debugging article we stepped into the B::C compiler with Od.
In the second B::C debugging article we stepped the c code with gdb.
We saw that the compiler is missing a sub in a seperate package, the most typical
problem with the current compiler. (since 1995)

We have a cool workaround for this, the -u option for "use".
-udummy forces all symbols in the package dummy to be dumped.

Remember, our testcase was:
'package dummy;sub meth{print "ok"};package main;dummy->meth' as ccode35.pl

$ perl -MO=C,-Do,-v,-udummy,-occode35.c ccode35.pl

Starting compile
Walking tree
walkoptree: 0. LISTOP (0x1471a20) leave
walkoptree: 1. OP (0x1471a48) enter
walkoptree: 1. COP (0x14719e0) nextstate
walkoptree: 1. UNOP (0x1471998) entersub
walkoptree: 2. OP (0x14719c0) pushmark
walkoptree: 2. SVOP (0x1472000) const
walkoptree: 2. SVOP (0x1471978) method_named
done main optree, walking symtable for extras
Prescan for unused subs
Saving unused subs
walkoptree: 0. UNOP (0x164a6c0) leavesub
walkoptree: 1. LISTOP (0x164a698) lineseq
walkoptree: 2. COP (0x1472a00) nextstate
walkoptree: 2. LISTOP (0x1472d48) print
walkoptree: 3. OP (0x1523598) pushmark
walkoptree: 3. SVOP (0x147fad8) const
save context:
curpad names:
curpad syms:
%INC and @INC:
amagic_generation = 1
Writing output
Total number of OPs processed: 13
NULLOP count: 0
Loaded Cwd
Loaded B
Loaded IO
Loaded Fcntl
Loaded B::C
ccode35.pl syntax OK

Here it is!
0. UNOP (0x164a6c0) leavesub
1. LISTOP (0x164a698) lineseq
2. COP (0x1472a00) nextstate
2. LISTOP (0x1472d48) print
3. OP (0x1523598) pushmark
3. SVOP (0x147fad8) const

is indeed our missing &dummy::meth

$ perl -MO=Concise,dummy::meth ccode35.pl


dummy::meth:
5 <1> leavesub[1 ref] K/REFC,1 ->(end)
- <@> lineseq KP ->5
1 <;> nextstate(dummy 1 ccode35.pl:1) v ->2
4 <@> print sK ->5
2 <0> pushmark s ->3
3 <$> const(PV "ok") s ->4

Any sub argument to B::Concise dumps the optree for this sub,
not just main.

So we still have the compiler bug that dummy->meth is not detected as it says
Prescan for unused subs
Saving unused subs

We'll have to fix the scanner which tries to detect which symbols are used
and which not. Why not just dump all symbols?
We would then also dump the compiler package and all its dependencies.
This will work, but it is bloating the resulting code and reducing the speedup
which we get by not having to parse the perl code again, dumping the optree
and just executing the optree.

To be continued at part 4 - fixing method_named.

Leave a comment

About Reini Urban

user-pic Working at cPanel on B::C (the perl-compiler), p2, types, parrot, B::Generate, cygwin perl and more guts (LLVM, jit, optimizations).