Ticket #112 (closed defect: wontfix)

Opened 6 years ago

Last modified 6 years ago

regress.pike -d0xTOP_MOVE ninestones:630 gives weird output

Reported by: alain Owned by: gnugo
Priority: normal Milestone: 3.7.10
Component: source Version:
Severity: trivial Keywords:
Cc: patch: yes

Description (last modified by alain) (diff)

regress.pike seems to do extra job after having given the answer. This does not seem to happen for others tests.

./regress.pike -v -e "/usr/local/bin/gnugo" \
                -o "--mode gtp -d0x400000" ninestones:630

Top moves:
1. F5  75.87
2. B11 72.31
3. G16 16.92
4. J14 14.21
5. O13 10.73
6. F18 8.75
7. G15 8.55
8. S13 7.99
9. H14 7.99
10. O15 7.82
ninestones:630  pass F5 [F5|H7]

Top moves:
1. A4  41.01
2. N9  21.23
3. H9  19.47
4. O7  13.95
5. C1  12.80
6. C8  11.20
7. C10 8.53
8. P11 7.00
9. K8  6.65
10. C12 6.37

Top moves:
1. E6  138.03
2. F7  133.35
3. H7  102.08
4. G7  91.73
5. H8  78.00
6. G8  73.90
7. J7  68.83
8. C1  42.20
9. C8  29.20
10. D1  28.97

Top moves:
1. C8  41.01
2. N9  21.23
3. H9  17.96
4. H8  17.33
5. H7  15.62
6. O7  13.95
7. D8  8.84
8. G8  8.82
9. G7  8.63
10. E10 8.08
ninestones                              15.22   3167312    7453    35437
Total nodes: 3167312 7453 35437
Total time: 15.22 (29.22)
Total uncertainty: 0.00

Attachments

ninestones.patch Download (584 bytes) - added by alain 6 years ago.
tiny patch. In the test file, some reg_genmove does not correspond to a test/answer, and are always executed, even if only one test is aked for.

Regression Results

Attachment Rev. PASS FAIL Nodes Status
ninestones.patch Download never tested

Change History

comment:1 Changed 6 years ago by alain

  • Description modified (diff)

comment:2 Changed 6 years ago by arend

To debug problems like this, use the 'gtp-dump-commands <file>' option of GNU Go.

What happens is that regress.pike is replaying all the reg_genmove commands preparing the ninestones:670 test later on. One could argue that regress.pike should stop as soon as it has reached the relevant line for the 630 test, but that's hardly an urgent problem.

Changed 6 years ago by alain

tiny patch. In the test file, some reg_genmove does not correspond to a test/answer, and are always executed, even if only one test is aked for.

comment:3 Changed 6 years ago by alain

  • Severity changed from normal to trivial
  • patch set

comment:4 Changed 6 years ago by gunnar

  • Status changed from new to closed
  • Resolution set to wontfix
  • Milestone changed from 3.8 to 3.7.10

I don't see any value in adding dummy tests. Regress.pike doesn't have any actual problem with unnumbered reg_gemoves. To see move values for a given test I recommend trying out view.pike instead.

The reason why regress.pike doesn't stop immediately after the 630 test command is that for testcases like trevor:262

loadsgf games/trevor/trevor_26.sgf 33
trymove black C1
trymove white C2
trymove black F1
262 defend C2
#? [0]
popgo
popgo
popgo

the popgo commands must be run to get the engine back into a proper state for (possible) later tests. The policy is that unnumbered commands are always run, to guarantee as well as possible against surprises.

Note: See TracTickets for help on using tickets.