This project is mirrored from https://github.com/haskell/Cabal. Pull mirroring updated .
  1. 06 Oct, 2016 13 commits
  2. 04 Oct, 2016 4 commits
    • Duncan Coutts's avatar
      Optimise Distribution.Version.mkVersion · 883b3c30
      Duncan Coutts authored
      883b3c30
    • Herbert Valerio Riedel's avatar
      Fix thinko in `decodeStringUtf8` · ed329074
      Herbert Valerio Riedel authored
      This resulted in some two-bytes utf8 encodings to be decoded
      into U+FFFD unintentionally (such as e.g. U+0142).
      
      With this fix, the property
      
          [ c | c <- [minBound..maxBound]
              , c < '\xD800' || c >= '\xE000' -- surrogate pair codes
              , (decodeStringUtf8 . encodeStringUtf8) [c] /= [c]
              ] == ['\xfffe','\xffff']
      
      holds. It's not clear to me why U+FFFE and U+FFFF ought to be singled
      out. Needs more investigation.
      
      TODO: testsuite coverage
      ed329074
    • Herbert Valerio Riedel's avatar
      Fix embarassing thinko in `encodeStringUtf8` · a87fcd10
      Herbert Valerio Riedel authored
      TODO: we need more unit-tests to catch this; also I noticed
      that `decodeStringUtf8` (which was previously named `fromUTF8BSImpl`)
      appears to have been broken for quite some time.
      a87fcd10
    • Herbert Valerio Riedel's avatar
      Add Data.Word/Int to Compat.Prelude · e14da30b
      Herbert Valerio Riedel authored
      For one this compensates for `Word` being exported from `Prelude` only with
      base >= 4.8, and moreover this brings the well-known {Int,Word}{8,16,32,64}
      names into scope by default when using the Compat.Prelude.
      e14da30b
  3. 03 Oct, 2016 1 commit
    • Herbert Valerio Riedel's avatar
      Optimise internal representation of `ModuleName` (#3927) · b0e958a3
      Herbert Valerio Riedel authored
      This changes the representation of module names from `[[Char]]`
      (e.g. `Control.Monad.Fail` is decomposed into `["Control","Monad","Fail"]`)
      which results in many small heap objects, to a strict list of `ShortText`s.
      
      `ModuleName` was already an opaque type, so there is no visible change
      to the exposed API.
      b0e958a3
  4. 02 Oct, 2016 5 commits
    • Herbert Valerio Riedel's avatar
      Make `AbiHash` opaque & `ShortText`-backed (#3921) · 7378066d
      Herbert Valerio Riedel authored
      This doesn't have any noteworthy performance impact since
      there aren't that many `AbiHash` values to keep track of and is
      rather for consistency with the other opaque types.
      7378066d
    • Herbert Valerio Riedel's avatar
    • Herbert Valerio Riedel's avatar
      Optimise `Ord Version` instance · aabee7c3
      Herbert Valerio Riedel authored
      Profiling showed that there are quite a few 'Ord' calls (~70k calls)
      which result in 'versionNumbers' being evaluated because the need to
      compare with Versions not fitting in a Word64 becomes necessary, mostly
      when inserting into a `Map`. So I've optimised this a bit more.  After
      some experimentation, the new `compare` implemetation results in faster
      comparisions (`cmpOpt` is the new optimised impl, `compare` is the
      previous naive one):
      
      comparing two PV1 constructors:
      
          benchmarking compare 1.2.3.4.5 1.2.3.4.5
          time                 42.33 ns   (42.17 ns .. 42.47 ns)
                               1.000 R²   (1.000 R² .. 1.000 R²)
          mean                 42.17 ns   (42.13 ns .. 42.28 ns)
          std dev              205.1 ps   (128.7 ps .. 325.7 ps)
      
          benchmarking cmpOpt  1.2.3.4.5 1.2.3.4.5
          time                 33.31 ns   (33.23 ns .. 33.40 ns)
                               1.000 R²   (1.000 R² .. 1.000 R²)
          mean                 33.37 ns   (33.29 ns .. 33.46 ns)
          std dev              288.6 ps   (242.9 ps .. 315.8 ps)
      
          benchmarking compare [1.2.3.4.5] [1.2.3.4.5]
          time                 31.92 ns   (31.89 ns .. 31.94 ns)
                               1.000 R²   (1.000 R² .. 1.000 R²)
          mean                 31.89 ns   (31.88 ns .. 31.91 ns)
          std dev              37.09 ps   (24.38 ps .. 54.82 ps)
      
      comparing a PV0 constructor with a PV1 constructor when the first digit
      decides the outcome:
      
          benchmarking compare 2.3.4 1.2.3.4.5
          time                 24.96 ns   (24.78 ns .. 25.25 ns)
                               0.996 R²   (0.989 R² .. 1.000 R²)
          mean                 25.50 ns   (24.95 ns .. 26.95 ns)
          std dev              2.802 ns   (1.275 ns .. 5.163 ns)
          variance introduced by outliers: 93% (severely inflated)
      
          benchmarking cmpOpt  2.3.4 1.2.3.4.5
          time                 11.29 ns   (11.27 ns .. 11.30 ns)
                               1.000 R²   (1.000 R² .. 1.000 R²)
          mean                 11.28 ns   (11.27 ns .. 11.29 ns)
          std dev              24.69 ps   (12.02 ps .. 46.86 ps)
      
          benchmarking compare [2.3.4] [1.2.3.4.5]
          time                 11.05 ns   (11.04 ns .. 11.06 ns)
                               1.000 R²   (1.000 R² .. 1.000 R²)
          mean                 11.05 ns   (11.04 ns .. 11.06 ns)
          std dev              32.82 ps   (15.89 ps .. 61.64 ps)
      
      These timings are now very close to the 'Ord [Int]' timings.
      
      With this optimisation, the total number of 'versionNumbers' calls
      reported by `+RTS -p` for `cabal new-build --dry`'ing haskell-ide-engine
      went down from originally 73501 calls to 6789 calls.
      aabee7c3
    • Herbert Valerio Riedel's avatar
      Optimise `Version` representation · cbddd8cb
      Herbert Valerio Riedel authored
      This optimises the `[Int]` representation to a 16-byte heap object for
      ~99% of version numbers (up to 4 components, each within a [0..0xfffe]
      value range) occuring on Hackage.
      
      One noteworthy improvement of this optimisation is a significant reduction
      of the size of the 01-index.cache file from previously 6299700 bytes
      (before #3905) down to 3408864 bytes, i.e. down to ~54%
      
      Also, this reduces the memory footprint and GC overhead a bit for
      e.g. `cabal info zzz` (which reads in the index cache) from
      
          cabal.0: There is no package named 'zzz'.
          You may need to run 'cabal update' to get the latest list of available
          packages.
               859,337,368 bytes allocated in the heap
               447,261,128 bytes copied during GC
                37,385,208 bytes maximum residency (19 sample(s))
                 1,311,136 bytes maximum slop
                       103 MB total memory in use (0 MB lost due to fragmentation)
      
                                               Tot time (elapsed)  Avg pause  Max pause
            Gen  0      1613 colls,     0 par    0.268s   0.268s     0.0002s    0.0012s
            Gen  1        19 colls,     0 par    0.227s   0.227s     0.0119s    0.0506s
      
            TASKS: 4 (1 bound, 3 peak workers (3 total), using -N1)
      
            SPARKS: 0 (0 converted, 0 overflowed, 0 dud, 0 GC'd, 0 fizzled)
      
            INIT    time    0.001s  (  0.001s elapsed)
            MUT     time    0.431s  (  0.758s elapsed)
            GC      time    0.495s  (  0.495s elapsed)
            EXIT    time    0.006s  (  0.005s elapsed)
            Total   time    0.934s  (  1.259s elapsed)
      
            Alloc rate    1,991,870,623 bytes per MUT second
      
            Productivity  46.9% of total user, 34.8% of total elapsed
      
      to
      
          cabal.1: There is no package named 'zzz'.
          You may need to run 'cabal update' to get the latest list of available
          packages.
               834,314,392 bytes allocated in the heap
               440,791,176 bytes copied during GC
                36,663,112 bytes maximum residency (19 sample(s))
                 2,225,040 bytes maximum slop
                        96 MB total memory in use (0 MB lost due to fragmentation)
      
                                               Tot time (elapsed)  Avg pause  Max pause
            Gen  0      1574 colls,     0 par    0.254s   0.254s     0.0002s    0.0007s
            Gen  1        19 colls,     0 par    0.223s   0.223s     0.0118s    0.0474s
      
            TASKS: 4 (1 bound, 3 peak workers (3 total), using -N1)
      
            SPARKS: 0 (0 converted, 0 overflowed, 0 dud, 0 GC'd, 0 fizzled)
      
            INIT    time    0.001s  (  0.001s elapsed)
            MUT     time    0.383s  (  0.699s elapsed)
            GC      time    0.477s  (  0.477s elapsed)
            EXIT    time    0.005s  (  0.005s elapsed)
            Total   time    0.869s  (  1.182s elapsed)
      
            Alloc rate    2,175,866,164 bytes per MUT second
      
            Productivity  44.9% of total user, 33.0% of total elapsed
      cbddd8cb
    • Edward Z. Yang's avatar
      GHC 8.1 compatibility fixes. · effb4ea6
      Edward Z. Yang authored
      
      
      It seems that GHC HEAD is now unable to infer the correct
      type for "f . g" when g has a higher-rank type.  So eta-expand
      all such occurrences to keep the build working.
      Signed-off-by: default avatarEdward Z. Yang <ezyang@cs.stanford.edu>
      effb4ea6
  5. 01 Oct, 2016 3 commits
  6. 30 Sep, 2016 2 commits
  7. 29 Sep, 2016 3 commits
    • Herbert Valerio Riedel's avatar
    • Herbert Valerio Riedel's avatar
      Move ShortText to Distribution.Utils.ShortText to avoid cycles · e204a346
      Herbert Valerio Riedel authored
      This moves `String`/UTF8 conversion helpers to Distribution.Utils.String
      e204a346
    • Duncan Coutts's avatar
      Adjust behaviour of reportBuildFailures · 4d308c04
      Duncan Coutts authored
      Make sure it consistently uses stderr, rather than a mixture of stdout
      and stderr. Also rename it to dieOnBuildFailure to make it clear that
      it is fatal in the case of build failures.
      
      In general we report build failures in two steps: header plus build log
      for each failing package, then a final die exception with summary.
      Previously the build log step was reported to stdout whereas the die
      exception message gets reported to stderr. So we switch that to use
      new dieMsg and dieMsgNoWrap utils so that it all goes to stderr. Also,
      like die, these are reported irrespective of the verbosity.
      
      This is more or less just a workaround for the fact that we do not yet
      have a nice structured/formatted error mechanism that would allow us to
      throw just the one error in this case.
      4d308c04
  8. 28 Sep, 2016 2 commits
    • Herbert Valerio Riedel's avatar
      Add `Distribution.Simple.Utils.ShortText` type (#3898) · 993d20a2
      Herbert Valerio Riedel authored
      This implements a type with a compact representation of `[Char]`.
      
      The data is stored internally as UTF8 in an 'Data.ByteString.Short.ShortByteString'
      when compiled against `bytestring >= 0.10.4`, and otherwise in a
      plain old `[Char]`.
      
      `ShortByteString` is available only from `bytestring` 0.10.4 on, and GHC
      7.8.4 was the first GHC to bundle `binary-0.10.4`. So this fallback
      affects mostly only GHC 7.6 and older.
      
      Note: Originally a strict `ByteString` was used as fallback for this patch. However, the 
      `[Char]` fallback avoids pinned memory and may be more preferable when dealing with
      many small `ShortText`s
      993d20a2
    • Herbert Valerio Riedel's avatar
      Make `Version` type opaque (#3905) · bb2026c4
      Herbert Valerio Riedel authored
      Similiar to dabd9d98 which made
      `PackageName` opaque, this makes `Distribution.Version.Version` opaque.
      
      The most common version numbers occuring on Hackage are 3- and
      4-part versions. This results in significant Heap overhead due to
      `Data.Version`'s inefficient internal representation.
      
      So like the `PackageName` commit, this commit is a preparatory commit to
      pave the way for replacing `Version`'s internal representation by a
      representation with a memory footprint which can be an order of
      magnitude smaller.
      
      Finally, this also adds a new functor-like convenience function
      
          alterVersion :: ([Int] -> [Int]) -> Version -> Version
      
      for modifying the version number components.
      bb2026c4
  9. 27 Sep, 2016 1 commit
    • Herbert Valerio Riedel's avatar
      Make `PackageName` type opaque (#3896) · dabd9d98
      Herbert Valerio Riedel authored
      When looking at heap-profiles of `cabal-install`, the `(:)` constructor
      stands out as the most-allocated constructor on the heap.
      
      Having to handle 10k+ package names contributes to the allocation
      numbers, especially on 64bit archs where ASCII `String`s have a 24 byte
      per character footprint.
      
      This commit is a preparatory commit to pave the way for changing
      `PackageName`'s internal representation to something like
      `ShortByteString` (which is essentially a thin wrapper around primitive
      `ByteArray#`s which themselves have have an overhead of 2 words + one
      byte per ASCII character rounded up to nearest word) which would allow
      to reduce the memory footprint by a full order of magnitude, as well as
      reduce pointer chasing and GC overhead.
      dabd9d98
  10. 26 Sep, 2016 4 commits
    • Herbert Valerio Riedel's avatar
      Introduce `Distribution.Compat.Prelude.Internal` + WARNING · 4c730f58
      Herbert Valerio Riedel authored
      This unexposes `Distribution.Compat.Prelude` again, and adds
      an exposed `.Internal` module which reexports the `.Prelude` module with
      an attached module-level `WARNING` pragma.
      
      This way users are discouraged to use this in `Setup.hs` files as they'd
      see compile warning like:
      
          Foo.hs:19:1-55: warning: [-Wdeprecations]
            Module ‘Distribution.Compat.Prelude.Internal’:
              This modules' API is not stable. Use at your own risk, or better yet, use @base-compat@!
      
      In `cabal-install` we import the `.Internal` module exactly once, and
      use -Wno-deprecations to suppress that the warning of that single
      import.
      4c730f58
    • Herbert Valerio Riedel's avatar
      Introduce `Distribution.Client.Compat.Prelude` · 82a22706
      Herbert Valerio Riedel authored
      This is supposed to become more or less a superset of Cabal's
      `Distribution.Compat.Prelude`.
      
      As a side-effect,t his exposes `Distribution.Compat.Prelude` from the
      Cabal library (which may be actually a good thing, as it may be useful
      module to Setup.hs writers).
      82a22706
    • Duncan Coutts's avatar
      Add InstallPlan.{toGraph,toMap,keys,keysSet} · 9b40b06c
      Duncan Coutts authored
      At least keysSet will be useful shortly. Others are just for good
      measure (though there's probably some cases where we could avoid
      converting to and from lists if we used toMap or toGraph).
      9b40b06c
    • Duncan Coutts's avatar
      Better errors when the InstallPlan invariant is violated · 12a39b68
      Duncan Coutts authored
      Same idea as a patch proposed by ezyang but with somewhat less
      infrastructure needed. We just use individual asserts for each part of
      the invariant, so the assert source location tells us which part was
      violated. Hopefully call stacks can also tell us in which function the
      invariant was checked.
      12a39b68
  11. 24 Sep, 2016 1 commit
  12. 22 Sep, 2016 1 commit