git.haldean.org ubik / 02a5277 docs / log
02a5277

Tree @02a5277 (Download .tar.gz)

log @02a5277raw · history · blame

.

20160315

@ 08:45
finally fixed the leak with native URIs, expeli works with leak sanitizer on now
made all the unit tests work with leak sanitizer on

@ 23:10
Debian stopped working, so got an arch vm running and everything installed,
newer gcc means better asan and more errors. A few of the unit tests didn't
work; those are fixed now, as are all of the pyasm tests. fib.xl compiles will
almost zero memory errors.

.

20160316

@ 18:30
Memory used by compiler is mostly all cleaned up! Unit tests and test progs run
with no leak errors except for the import test, which is use-after-free-ing. No
luck yet on figuring out why; for some reason URIs at package scope end up with
one-too-few references so when the dependency record gets cleaned up the URI is
freed.

Got ctags working (it was really easy).

@ 23:00

Fixed the use-after-free, just one leak left on import.xl... and it's fixed!
"make check" finally passes with the memory sanitizer on. Thank fuck.

.

20160317

@ 19:00

Discovered that the debugging utilities I added to the GC module are actually
retaining pointers to things that should be identified as leaked. The leak hunt
is not over.

.

20160320

@ 17:45

Spent a while working on memory problems. Discovered that failures in the pyasm
checks had been masking failures in the test/prog checks. Fixed the test/prog
checks, but still stumped on the pyasm memory leaks.

Added please-leak.xlpy, which exercises the leak in the pyasm-generated graphs.
Interestingly, the leak only occurs in this specific graph when the terminal
node appears twice. More generally, it looks like the leaked graph is always the
graph corresponding to the function that is called by the modinit.