Crash on SPARC

[bakpakin]

Crash on SPARC

[tankf33der]

Now janet can pass all tests on test files 1,2,3,5.
Huge progress.
For failed tests on suite[0,4].janet I will open new issue.

[tankf33der]

pulled latest. recompiled.

Notes:

1. All 112 tests passed from suite2.janet
2. Bus error from suite3.janet, I cut related code:

(var num-tests-passed 0)
(var numchecks 0)
(var num-tests-run 0)

(defn assert [x e]
 (++ num-tests-run)
 (when x (++ num-tests-passed))
 (if x
   (do
     (when (= numchecks 25)
       (set numchecks 0)
       (print))
     (++ numchecks)
     (file/write stdout "\e[32m?\e[0m"))
   (do
     (file/write stdout "\n\e[31m?\e[0m  ")
     (set numchecks 0)
     (print e)))
 x)

(defn check-deep
  [pat text what]
  (def result (peg/match pat text))
  (assert (deep= result what) text))

(def csv
  '{:field (+
            (* `"` (% (any (+ (<- (if-not `"` 1)) (* (constant `"`) `""`)))) `"`)
            (<- (any (if-not (set ",\n") 1))))
    :main (* :field (any (* "," :field)) (+ "\n" -1))})

(defn check-csv
  [str res]
  (check-deep csv str res))

(check-csv "1,\"2\",3" @["1" "2" "3"])

New core print:

core 'core' of 7053:    ./build/janet tt.janet
 0000000100037e24 peg_rule (1001a36dc, 1001ad8f8, 1001a36dc, 1000253e8, 100037600, ffffffff7ffff4a0) + 650
 0000000100037bbc peg_rule (1001ad8f0, 1001ad8f4, 1001a36dc, 1000253e8, 100037600, ffffffff7ffff4a0) + 3e8
 0000000100037d20 peg_rule (1001a36dc, 1001ad8b8, 1001a36dc, ffffffff, 1, ffffffff7ffff4a0) + 54c
 0000000100038170 peg_rule (ffffffff7ffff4a0, 1001ad89c, 1001a36db, 1, 0, ffffffff7ffff4a0) + 99c
 0000000100037bbc peg_rule (1001ad888, 1001ad88c, 1001a36db, 1000253e8, 100037600, ffffffff7ffff4a0) + 3e8
 0000000100037b00 peg_rule (2, 1001ad878, 1001a36da, 1000253e8, 100037600, ffffffff7ffff4a0) + 32c
 0000000100037d20 peg_rule (1001a36d9, 1001ad96c, 1001a36d9, ffffffff, 0, ffffffff7ffff4a0) + 54c
 0000000100037bbc peg_rule (1001ad864, 1001ad868, 1001a36d9, 1000253e8, 100037600, ffffffff7ffff4a0) + 3e8
 0000000100039cd0 cfun_peg_match (2, 1001ac9e8, 1001a36d8, 1001ad840, 7, 1001a36df) + cc
 0000000100044ddc run_vm (1001ac8e0, ffffffff7f5c2a6c, ffffffff7f5c2a70, 100043178, 2, 1001ab61c) + 1790
 000000010004362c janet_continue (1, ffffffff7f5c2aa8, ffffffff7ffff810, fffffffffffffff0, ffffffffffffffe0, ffffffffffffffe8) + 2fc
 0000000100043e54 run_vm (1001a7690, ffffffff7f5c2a6c, ffffffff7f5c2a70, 100043178, 1001aa7d0, 10016f104) + 808
 000000010004362c janet_continue (1, ffffffff7f5c2aa8, ffffffff7ffffa38, fffffffffffffff0, ffffffffffffffe0, ffffffffffffffe8) + 2fc
 000000010003b9f0 janet_dobytes (100150e20, 1001a7690, bbb, 100025fd8, ffffffff7ffffa68, 0) + e8
 0000000100046d30 main (2, ffffffff7ffffc48, 100150e20, 100178a60, 100046b9c, 100025fd0) + f8
 0000000100026064 _start (0, 0, 0, 0, 0, 0) + 64

[bakpakin]

Ok, so that's an improvement. Maybe add the -g flag to the Makefile CFLAGS so we can see which exact line is the issue (although I'm not sure if SPARC will provide that info). peg_rule is the function that interprets peg grammars at runtime, so is basically a big switch case. It would be helpful to know which rule caused the issue.

I also believe that the cause of the crash either has to do with alignment issues from pointer arithmetic that assumed too much, and less likely assuming little-endian.

[tankf33der]

the same with -g:

# pstack ./core 
core './core' of 15772: ./build/janet tt.janet
 0000000100037e24 peg_rule (1001a36dc, 1001ad8f8, 1001a36dc, 100025360, 100037600, ffffffff7ffff3e0) + 650 (peg.c:298)
 0000000100037bbc peg_rule (1001ad8f0, 1001ad8f4, 1001a36dc, 100025360, 100037600, ffffffff7ffff3e0) + 3e8 (peg.c:212)
 0000000100037d20 peg_rule (1001a36dc, 1001ad8b8, 1001a36dc, ffffffff, 1, ffffffff7ffff3e0) + 54c (peg.c:255)
 0000000100038170 peg_rule (ffffffff7ffff3e0, 1001ad89c, 1001a36db, 1, 0, ffffffff7ffff3e0) + 99c (peg.c:332)
 0000000100037bbc peg_rule (1001ad888, 1001ad88c, 1001a36db, 100025360, 100037600, ffffffff7ffff3e0) + 3e8 (peg.c:212)
 0000000100037b00 peg_rule (2, 1001ad878, 1001a36da, 100025360, 100037600, ffffffff7ffff3e0) + 32c (peg.c:194)
 0000000100037d20 peg_rule (1001a36d9, 1001ad96c, 1001a36d9, ffffffff, 0, ffffffff7ffff3e0) + 54c (peg.c:255)
 0000000100037bbc peg_rule (1001ad864, 1001ad868, 1001a36d9, 100025360, 100037600, ffffffff7ffff3e0) + 3e8 (peg.c:212)
 0000000100039cd0 cfun_peg_match (2, 1001ac9e8, 1001a36d8, 1001ad840, 7, 1001a36df) + cc (peg.c:1061)
 0000000100044ddc run_vm (1001ac8e0, ffffffff7f5c2a6c, ffffffff7f5c2a70, 100043178, 2, 1001ab61c) + 1790 (vm.c:578)
 000000010004362c janet_continue (1, ffffffff7f5c2aa8, ffffffff7ffff750, fffffffffffffff0, ffffffffffffffe0, ffffffffffffffe8) + 2fc (vm.c:832)
 0000000100043e54 run_vm (1001a7690, ffffffff7f5c2a6c, ffffffff7f5c2a70, 100043178, 1001aa7d0, 10016f104) + 808 (vm.c:636)
 000000010004362c janet_continue (1, ffffffff7f5c2aa8, ffffffff7ffff978, fffffffffffffff0, ffffffffffffffe0, ffffffffffffffe8) + 2fc (vm.c:832)
 000000010003b9f0 janet_dobytes (100150e20, 1001a7690, bbb, 100025a60, ffffffff7ffff9a8, 0) + e8 (run.c:50)
 0000000100046d30 main (2, ffffffff7ffffb98, 100150e20, 100178a60, 100046b9c, 100025fe0) + f8 (main.c:50)
 0000000100026064 _start (0, 0, 0, 0, 0, 0) + 64

[bakpakin]

Ah, I think I can guess the culprit - probably the make_peg function that is doing pointer arithmetic to pack multiple arrays into one malloc (the peg->constants array was, I guess, not correctly aligned). Some changes in 9495be3 should fix the bus error (hopefully).

[tankf33der]

I will try again in 7days, im on vacation now. Sorry for delay.