methodtailcall broken

[kjs]

methodtailcalls are broken. The next example shows this.
the call to foo() fails, whereas the semantic equivalent call to bar() is ok. foo() returns with a method tailcall, bar() returns with a normal return statement.


.sub main
    say "ok 1"
    $P0 = foo() ## fails :-(
    $P0 = bar() ## ok.
    say "ok 2"
    $I0 = $P0
    say "ok 3"
    say $I0
    say "ok 4"
.end

.sub foo
    .local pmc p
    p = new "Class"
    .tailcall p."attributes"()
.end


.sub bar
    .local pmc  p
    p = new "Class"
    $P0 = p."attributes"()
    .return ($P0)
.end

[whiteknight]

Don't method tailcalls use the methodtailcall opcode instead of the tailcall one? Does the .tailcall directive properly recognize this and use the appropriate opcodes for each?

[whiteknight]

I looked this up in IMCC and the code appears to trace out correctly: IMCC is identifying the tailcall as a method tailcall and is generating a tailcallmethod op to handle it. It appears that method tailcalls are used in several places during the build, but I haven't had the time to look to see how any of those cases are different from the test case presented here.

[coke]

This test passes for me in r38816.

[coke]

whoops. The test generates "ok 1\n" and stops; I missed additional output was expected; This sample should be added as a test.

[NotFound]

Testing with r45981 in debian/amd64:

ok 1
ok 2
ok 3
0
ok 4

Is this the expected result?

[bacek]

Hello.

NotFound, yes, this is expected result. Test added at r46468. Resolving ticket.

-- Bacek