Shrinking is Insanely Slow for write_read

[BebeSparkelSparkel]

Just one test took ~25 minutes due to shrinking. https://github.com/haskell/text/actions/runs/9201285542/job/25309228763?pr=590#step:6:1224

Should noShrinking be added?

[Bodigrim]

I imagine at the very least we don't want to shrink first three arguments of

  propTest :: TextEncoding -> Gen IO.BufferMode -> IO.NewlineMode -> c -> Property

[BebeSparkelSparkel]

The encoding is not shrunk. I have wrapped the other two with NoShrink but without much improvement. See #597

[Bodigrim]

      input-output
        t_write_read
          UTF-8
            NoBuffering:                                      OK (0.80s)
              +++ OK, passed 100 tests; 39 discarded.
            LineBuffering:                                    OK (0.10s)
              +++ OK, passed 100 tests; 27 discarded.
            BlockBuffering:                                   FAIL (1448.46s)
              *** Failed! (after 72 tests and 2007 shrinks):
              Exception:
                Assertion failed
                CallStack (from HasCallStack):
                  assert, called at src/Data/Text/IO.hs:249:35 in text-2.1.1-inplace:Data.Text.IO
              NewlineMode {inputNL = LF, outputNL = LF}
              ["aaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaa","aaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","a","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaa","aa","aaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaa","aa","","aaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"]
              BlockBuffering (Just 4)
              Use --quickcheck-replay=934323 to reproduce.
              Use -p '/t_write_read.UTF-8.BlockBuffering/' to rerun this test only.

2007 shrinks is a high number, but on the other hand why did they take 1448 seconds? That's 0.7 seconds per test, which sounds a lot, given that the test data does not seem to be that huge.

[BebeSparkelSparkel]

I think many more tests are run than 2007. If you look at numShrinkTries, so I think 2007 represents numShrinks but the number of tries is not shown.

[BebeSparkelSparkel]

I imagine there are so many shrinks tried because there are many variables to shrink

• length of list
• length of each text
• every character

I think if the character variable is removed this would perform much better. Either, do not modify the characters or set them all to 'a'.

[Bodigrim]

Dunno, maybe lumping NoShrink is the simplest solution. I'm not too keen to have a custom wrapper with instance Arbitrary here...

[BebeSparkelSparkel]

I just added NoShrink to quickcheck and it has been pulled nick8325/quickcheck#408. We could up the version of quickcheck and use it.