libera/#clasp - IRC Chatlog
Search
14:22:15
bike
symbols for symbol macros might be nice, but they wouldn't be very good in a hash table
14:24:41
bike
symbol macros and conses are the only forms that can end up actually signaling errors, so i think that's about all we need to worry about. technically an asynchronous interrupt could end up anywhere, but i don't think we need to localize the position to a particular self evaluating constant or variable lookup
14:25:58
yitzi
and use that as the key in the hashtable. Then the bytecode compiler can just count out the same indicies
14:38:29
bike
the compile-4a failure is just because of the describe-form thing. should i fix that by editing the history or do a new commit?
16:06:25
bike
is there any chance the link problem is related to my increasing the stack size? i have no particular reason to believe that other than that it's a weird low level change, but nonetheless
16:51:03
bike
that's a little concerning. maybe it would eventually be good to est u pthe stack with anonymous mmaps or something, along with setting up a proper guard page
18:24:12
yitzi
They have "upgrade" tests too. I skipped those on CI cause they are slow and we ship ASDF so "upgrading" is pretty silly. Plus they are slow as f--k.
18:26:55
yitzi
Yeah, its a run-program test failure. I think there is a bug lurking in our run-program stuff. I get segmentation faults sometimes.
18:31:33
yitzi
If the ASDF people were running a recent Clasp version I'd feel better about skipping the tests. They are using v0.4.3. I submitted a PR the cl image repo to update to v2.3.0 but there doesn't really seem to be a lot of activity there. Plus they have a different docker image for every implementation. Its kind of a heavy-weight solution. If I was trying to do the equivalent I'd make a single docker image based on Arch with all the implementations