libera/#clasp - IRC Chatlog

10:30:28 yitzi beep

10:30:29 Colleen yitzi: drmeister said 7 hours, 21 minutes ago: - I may have uncovered a bug in your LOOP macro? Or a bug in a system called :CHUNGA that Hunchentoot relies on. Claude monkey patched a function and now Huntchentoot appears to work.

10:30:39 drmeister Hi

10:30:43 yitzi Hello

10:31:17 yitzi I can "fix" Khazern, bit a couple of things about LOOP first....

10:32:05 drmeister I'm not sure there's actually a problem. I was moving pretty fast last night trying to get hunchentoot working in Cando and that came up.

10:32:14 drmeister I posted about it and then kept going.

10:32:27 yitzi The initial values of all LOOP variables in the spec is unspecified. SBCL is MIT LOOP as is most CL implementations.

10:32:38 drmeister But that monkey patch for chunga:trim-whitespace seems to fix the problem.

10:32:49 yitzi MIT LOOP does some pretty questionable stuff with LOOP variables.

10:33:43 drmeister Ok - and sbcl uses MIT LOOP and chunga has an implementation specific bug according to the clhs?

10:33:53 yitzi Khazern initializes LOOP variables to default values and only sets them to step values if the LOOP actually runs.

10:33:54 drmeister It's a story as old as time.

10:35:27 yitzi For example, (LOOP FOR I OF-TYPE INTEGER IN '() FINALLY (RETURN I)) in Khazern returns 0 even though the LOOP never runs cause I is an integer.

10:36:26 yitzi But for your example, the loop variable has an implied type of INTEGER and since the LOOP doesn't run the default value of INTEGER is 0.

10:36:51 yitzi That has been the strict interpretation of Khazern.

10:36:54 yitzi That said...

10:37:50 yitzi To me a loop of (LOOP FOR I DOWNFROM 0 TO 0 ...) implies that value is I is in the interval [0.0]

10:38:35 yitzi And (LOOP FOR I DOWNFROM -1 TO -1 ...) implies that the value of I is in the interval [-1.-1]

10:38:57 yitzi which would mean that the default value should be in the implied interval.

10:40:06 yitzi So I've made that change to Khazern and Clasp has PR https://github.com/clasp-developers/clasp/pull/1759

10:41:17 yitzi On a side note. I think that is stuff in Clasp or Cando that you haven't pushed. https://p.blueridgedebate.com/?fc7256b76cc049d9#F4N39eZYcRzwbpUVy9N1LrfK8N4jCqBzaLxVkSpfx7pB

10:42:24 yitzi drmeister: ^

10:46:59 drmeister I'll push everything I have.

11:17:32 drmeister I think everything is pushed and it builds on my system. If that problem persists then I'll have to dig deeper.

11:37:54 yitzi I am suspicious about that :external-format change. :default is actually the default external-format. Not NIL.

11:39:47 yitzi I think that sockets.lisp is correct, but our C code is probably doing something wrong with :external-format :default.

11:43:35 yitzi bike: we have low-level bivalent streams. I added them a while back. All you have to do is change the stream element type, via (setf (stream-element-type stream) 'integer) or (setf (stream-element-type stream) 'character)

11:43:39 yitzi drmeister: ^

11:45:24 drmeister Ok. I'll check with bike.

11:47:18 yitzi I suspect that hunchentoot is just assuming you can do whatever you want without explicitly switching. Which wouldn't make a lot sense since in the case of character streams you also need the external-format to decide on the octet encoding.

11:48:18 yitzi The element-type and external-format are a pair that determines the reading & writing in other words.