I regularly run the test suite ("cabal test") to make sure my
re-factorings don't break anything.

I have previously reported failures in the unit tests for the
experimental V3 patch format, and it was suggested that these tests
should perhaps be disabled, because the V3 format is not currently used.

But now I got a failure for V2 patch format:

Darcs.Patch.V2 (using Prim.V1):
  [...]
  commute (tree): recommute: [Failed]
*** Failed! (after 77 tests): 
failed
conflictor [
rmdir ./v
]
:
adddir ./v/ce
:>
duplicate
rotcilfnoc [
adddir ./v/ce
]
:
rmdir ./v
addfile ./v/C.txt
:
rmfile ./v/C.txt
Sealed (WithStartState V1Model{ "v" "v/C.txt"[] } (ParTree (SeqTree (FP
(fp2fn "./v/C.txt") RmFile) (SeqTree (DP (fp2fn "./v") RmDir) NilTree))
(SeqTree (FP (fp2fn "./v/C.txt") RmFile) (SeqTree (DP (fp2fn "./v/ce")
AddDir) NilTree))))
(used seed 5231146206818707969)

Attaching the "cabal test" log.

Attachments

• darcs-2.9.9-darcs-test.log

It happened again:

Darcs.Patch.V2 (using Prim.V1):
  [...]
  commute (tree): nontrivial recommute: [Failed]
*** Failed! (after 94 tests): 
failed
conflictor {{
:
move ./EoM.txt ./Sp/HSa.txt
:
move ./Qfe.txt ./Sp/W.txt
:
rmdir ./Sp
}} []
:
move ./Qfe.txt ./iox.txt
:>
duplicate
rotcilfnoc {{
:
move ./Qfe.txt ./iox.txt
:
move ./Qfe.txt ./Sp/W.txt
:
rmdir ./Sp
}} []
:
move ./EoM.txt ./Sp/HSa.txt
rotcilfnoc {{
:
move ./Qfe.txt ./iox.txt
:
move ./Qfe.txt ./Sp/W.txt
}} []
:
rmdir ./Sp
rotcilfnoc [
move ./Qfe.txt ./iox.txt
]
:
move ./Qfe.txt ./Sp/W.txt
:
rmdir ./Sp
Sealed (WithStartState V1Model{ "EoM.txt"["E" "y" "t" "D" "M s" "d E"
"o" "H n" "q" "Z X" "I" "s" "d" "y" "V z" "D" "y L"] "Qfe.txt"["w" "s"
"h" "K" "g" "M" "W V" "F" "a" "Y c" "N" "q w" "Z y" "i" "e y" "I" "k"
"s" "e" "y r" "B y"] "Sp" } (ParTree (ParTree (SeqTree (FP (fp2fn
"./Qfe.txt") (Hunk 10 ["Y c","N","q w","Z y"] ["n"])) (SeqTree (DP
(fp2fn "./Sp") RmDir) NilTree)) (ParTree (ParTree (SeqTree (FP (fp2fn
"./EoM.txt") (TokReplace "A-Za-z_0-9" "o" "U")) NilTree) (SeqTree (Move
(fp2fn "./EoM.txt") (fp2fn "./Sp/HSa.txt")) NilTree)) (SeqTree (FP
(fp2fn "./Qfe.txt") (Hunk 18 ["s","e"] ["K S"])) (SeqTree (FP (fp2fn
"./q.txt") AddFile) NilTree)))) (ParTree (ParTree (SeqTree (Move (fp2fn
"./Qfe.txt") (fp2fn "./Sp/W.txt")) NilTree) (SeqTree (FP (fp2fn
"./EoM.txt") (Hunk 11 (packStringLetters "Isdy") (packStringLetters
"w"))) (SeqTree (DP (fp2fn "./Sp") RmDir) NilTree))) (SeqTree (DP (fp2fn
"./Sp") RmDir) (SeqTree (Move (fp2fn "./Qfe.txt") (fp2fn "./iox.txt"))
NilTree)))))
(used seed -7363009043839118403)

I am marking this a critical for now, as this is a bug in the very core
of the Darcs patch formalism. If it turns out that the property in
question doesn't actually need hold we can downgrade the priority.

Yeah, this is known - duplicates and conflictors are incorrectly
implemented, see http://bugs.darcs.net/issue2047 for a simpler example,
where the information loss that causes the issue is evident.

The trouble is, no one understands the Conflictor code, and in
particular no one has an understanding of where this issue comes from.
Indeed, it's going to need a new patch format to fix regardless
(existing repo behaviour could change due to the "fix").