William Coleda <[EMAIL PROTECTED]> wrote: > Still some GC bugs kicking around that neither Leo nor I have been > able to find yet. (Note that the bug in [string] may actually be a PGE > leak, as that code uses the glob rules).
Well there are GC and other bugs, probably. I get really strange results when running the tcl tests: $ make test Failed Test Stat Wstat Total Fail Failed List of Failed --------------------------------------------------------------- t/cmd_continue.t 1 256 2 1 50.00% 2 *but* I see much more failing: some of cmd_concat, cmd_insert, cmd_proc, and cmd_string, which aren't shown in the summary. Turning off DOD/GC gives this result: $ TEST_PROG_ARGS=-G make test All tests successful. Files=36, Tests=228, 43 wallclock secs (35.93 cusr + 6.15 csys = 42.08 CPU) again with above failing tests unnoticed. Running with -G didn't flush output w/o newlines properly, which is fixed now. Some more remarks: lib/tclword.imc: .local pmc me me = interpinfo 16 #INTERPINFO_CURRENT_OBJECT setattribute me, "TclWord\x00chunks", emptyArray this is extremly prone to errors. We don't guarantee that the constant 16 stays the current object. Just use C<self> here, which does the same: setattribute self, "TclWord\x00chunks", emptyArray tcl.imc: $P2 = parser."parse"(input_line_PMC,zero,zero) (retcode,retval) = interpret($P1) $P1 / $P2 seems to be a typo, which happens to work as it's likely that the register allocator will use the same register. leo
