October 2006 - Cvs-nhc98

patch applied (hat): Avoid unnecessary trace info for untraced code.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Fri Feb 1 11:33:49 PST 2002 olaf * Avoid unnecessary trace info for untraced code. Changed combinator ulazySat so that it no longer forces evaluation of the trace so that much of untraced code is actually traced. Reduces trace size substantially: Adjoxo from 63 MB to 41 MB; everything reachable according to hat-check. A trivial test program from 4 KB to 0.5 KB; nearly everything reachable according to hat-check; unreachable part needed by hat-detect. Reduces runtime for ghc; Adjoxo from 29s to 17s. However, increases runtime for nhc98; Adjoxo from 31s to 1m 45s. A time profile exposes the problem for nhc98. The new ulazySat uses a new IORef for every invocation. Especially the IO machinery and the unsafePerformIO calls are very expensive in nhc98. Some percentages of Adjoxo: >>= 5.4%; _mkIOok2 3.4%, _mkIOok3 1.6%, unsafePerformIO 5.3%, readIOArray 2.4%, writeIOArray 2.1%, newIOArray 1.9%, newIORef 1.1%, readIORef 0.5%, writeIORef 0.4%, Ix 12.2%, Array 4.6%; sum of these: 41% ulazySat itself 4% I'm still not sure that the new definition of ulazySat doesn't lose necessary trace part. Actually a trace should always be demanded before its value. A little modification of uLazySat however proved, that it is not always evaluated that way. uLazySat always returns the hidden parent when its trace is demanded after its normal value. M ./src/compiler98/Flags.hs -1 +1 M ./src/compiler98/HatTrans.hs -10 +3 M ./src/compiler98/TraceTrans.hs -3 +7 M ./src/hatlib/Hat.hs -5 +29 M ./src/hatlib/Makefile -1 +2 M ./src/hattrans/Flags.hs -1 +1 M ./src/hattrans/HatTrans.hs -10 +3 M ./src/hattrans/TraceTrans.hs -3 +7

1 0

patch applied (hat): The predicate `isTidCon' needs to check for tuples and the [] constructor
by Malcolm Wallace 10 Oct '06

10 Oct '06

Fri Feb 1 06:40:08 PST 2002 malcolm * The predicate `isTidCon' needs to check for tuples and the [] constructor in addition to upper-case- and colon- prefixed names. M ./src/compiler98/TokenId.hs -2 +3

1 0

patch applied (hat): Another fix for import/export of named fields. When a datatype is
by Malcolm Wallace 10 Oct '06

10 Oct '06

Fri Feb 1 06:15:38 PST 2002 malcolm * Another fix for import/export of named fields. When a datatype is exported abstract but its named fields are exported separately as vars, previously the field names were not being exported correctly. M ./src/compiler98/RenameLib.hs -20 +23

1 0

patch applied (hat): When creating the .hx file, forgot to check whether a field name is
by Malcolm Wallace 10 Oct '06

10 Oct '06

Fri Feb 1 03:49:50 PST 2002 malcolm * When creating the .hx file, forgot to check whether a field name is listed amongst the ordinary exports (as a variable), rather than just in the subordinate list under a type. M ./src/compiler98/AuxFile.hs +1 M ./src/hattrans/AuxFile.hs +1

1 0

patch applied (hat): Parse import/export of named fields more correctly.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Jan 31 08:41:39 PST 2002 malcolm * Parse import/export of named fields more correctly. * The compiler previously rejected export/import of a mixture of explicit field names and constructor names as subordinates of a type name. Now we accept it. * The import/export decl TC(a,b,c...) is ambiguous, because it could be a class with methods, or a type with only field names mentioned. Thus, there is no point in distinguishing them in the abstract syntax - they are sorted out in a later analysis anyway, in exactly the same way as the TC(..) form. * The .hx file mechanism now knows about field names and records them. M ./src/compiler98/AuxFile.hs -67 +74 M ./src/compiler98/AuxLabelAST.hs -4 +6 M ./src/compiler98/Need.hs -13 +13 M ./src/compiler98/Parse2.hs -6 +6 M ./src/compiler98/PreImport.hs -12 +16 M ./src/compiler98/PrettySyntax.hs -5 +2 M ./src/compiler98/Syntax.hs -4 +4 M ./src/compiler98/SyntaxPos.hs -4 +3 M ./src/compiler98/TokenId.hs -12 +8 M ./src/compiler98/TraceTrans.hs -10 +16 M ./src/hattrans/AuxFile.hs -67 +74 M ./src/hattrans/AuxLabelAST.hs -4 +6 M ./src/hattrans/TraceTrans.hs -10 +16

1 0

patch applied (hat): Re-instate (and fix) the explicit import/export lists for Flags fieldnames.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Jan 31 08:31:24 PST 2002 malcolm * Re-instate (and fix) the explicit import/export lists for Flags fieldnames. M ./src/compiler98/Flags.hs +1 M ./src/compiler98/HatTrans.hs -4 +2 M ./src/compiler98/MainNew.hs -8 +7 M ./src/hattrans/Flags.hs +1 M ./src/hattrans/HatTrans.hs -4 +2

1 0

patch applied (hat): Add function mkParLhs as a helper for parsing definitions of the form
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Jan 31 03:20:36 PST 2002 malcolm * Add function mkParLhs as a helper for parsing definitions of the form (a `f` b) c = ... M ./src/compiler98/MkSyntax.hs -1 +2

1 0

patch applied (hat): Reformat to 80-columns.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Jan 31 03:19:16 PST 2002 malcolm * Reformat to 80-columns. M ./src/compiler98/Parse.hs -35 +48

1 0

patch applied (hat): Remove illegal comma in export list.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Jan 31 02:24:52 PST 2002 malcolm * Remove illegal comma in export list. M ./src/compiler98/Flags.hs -1 +1 M ./src/hattrans/Flags.hs -1 +1

1 0

patch applied (hat): Second step towards allowing records in portable transformation.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Wed Jan 30 09:29:27 PST 2002 olaf * Second step towards allowing records in portable transformation. Transformation creates code for field selectors. However, this cannot yet be used, because the hx-renamer ignores field names and a field name in an expression hence leads to a runtime error. M ./src/compiler98/TraceTrans.hs -4 +58 M ./src/hattrans/TraceTrans.hs -4 +58

1 0