Repository navigation
Fusion not working as advertised #202
Description
Activity
- added a commit that references this issue
on Nov 7, 2017 I did a more comprehensive test of testing properties of the
textlibrary, and the test harness that I created spits out this information:Data.Text: Total functions (95) Untestable functions (32) Fuses (40): pack, decodeUtf8, singleton, unfoldr, unfoldrN, cons, snoc, append, tail, init, map, intersperse, toCaseFold, toLower, toUpper, toTitle, justifyLeft, scanl, take, drop, takeWhile, dropWhile, filter, unpack, head, last, null, compareLength, foldl, foldl', foldl1, foldl1', foldr, foldr1, any, all, maximum, minimum, findIndex, isPrefixOf Does not fuse (but should) (11): unpackCString#, reverse, scanl1, scanr1, takeWhileEnd, dropWhileEnd, dropAround, strip, stripStart, stripEnd, length Fuses (but should not) (2): find, index Does not fuse (10) Data.Text.Lazy: Total functions (100) Untestable functions (37) Fuses (36): pack, singleton, cons, snoc, append, tail, init, map, intersperse, toLower, toUpper, toTitle, justifyLeft, scanl, take, drop, takeWhile, dropWhile, filter, unpack, head, last, null, length, compareLength, foldl, foldl', foldl1, foldl1', foldr, foldr1, any, all, maximum, minimum, isPrefixOf Does not fuse (but should) (6): decodeUtf8, toCaseFold, scanl1, dropAround, strip, stripStart Fuses (but should not) (4): unfoldr, unfoldrN, find, index Does not fuse (17)- The
Untestablefunctions are functions that have additionalTextarguments that are not expected to fuse, but which would confuse my test of whether things fuse. But none of these claim to be “Subject to fusion” with the exception ofuncons, where the documentation claims that fusion happens, but the definitions is clearly not set up for fusion. - Yay, many functions that we claim fuse actually fuse!
- Most interesting are of course the
Does not fuse (but should)category. It includedlength, responsible for the original bug report. This is where the most work is needed. Fuses (but should not)means that the documentation for these functions shoul say “Subject to fusion”.
For most functions, the question of whether it fuses or not is a clear-cut yes/no question. An exception is
singleton: When a text pipeline starts withsingletonthen some functions suddenly fuse that did not before (e.g.center), others fail to fuse. I am not worried about that.This is tested with GHC-8.4 and text ==1.2.3.0.
Reacted by Simon Meier and rnd- The
- added a commit that references this issue
on Jun 21, 2019 Automatic fusion has been disabled some time ago.
The documentation of
Data.Textstatesbut (with GHC-8.0 or 8.2 and text-1.2.2.2) this is not true: Looking at
-ddump-simplI clearly see aTextvalue constructed of which the length is calculated.I assume that originally, the example from the introduction did fuse properly, and then it broke…
Do you have a test suite that somehow tests the fusion promises of
text? I stumbled over this because I am thinking about a proper, structured way of testing such non-functional properties of code in test suites, based on first successes with usingghc-proofsfor that ingeneric-lens, and maybetextwould be a good second case study.