- Cvs-nhc98 - Haskell.org

patch applied (hat): Extra configuration for IBM Power3-AIX architecture.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Wed Apr 25 06:08:22 PDT 2001 malcolm * Extra configuration for IBM Power3-AIX architecture. M ./script/harch +8

1 0

patch applied (hat): Corrected bug of unevaluated partial applications of constructors.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Wed Apr 25 03:31:55 PDT 2001 olaf * Corrected bug of unevaluated partial applications of constructors. The combinator `eagerSat' was not the source of the bug. Hence I resurrected it. The source was that data constructors defined in trusted modules were trusted. However, they should always be suspected so that their application is recorded in the trace. Changed fileformat.c accordingly. The Mate trace file grows by 2%. M ./src/compiler98/DbgTrans.hs -1 +1 M ./src/old/runtime/fileformat.c -1 +1

1 0

patch applied (hat): Remove debugging printf from signum (tracing only).
by Malcolm Wallace 10 Oct '06

10 Oct '06

Tue Apr 24 04:26:09 PDT 2001 malcolm * Remove debugging printf from signum (tracing only). M ./src/old/runtime/numeric.c -1 +1

1 0

patch applied (hat): A bunch of minor docmentation updates.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Tue Apr 24 02:48:27 PDT 2001 malcolm * A bunch of minor docmentation updates. M ./COPYRIGHT -9 +21 M ./docs/feature-table.html -12 +12

1 0

patch applied (hat): Allow hat-connect to find src files by following directories.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Fri Apr 20 07:52:18 PDT 2001 malcolm * Allow hat-connect to find src files by following directories. M ./src/hattools/utils.c -6 +39

1 0

patch applied (hat): Enable sharing of traced methods that are cafs.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Apr 19 07:26:13 PDT 2001 olaf * Enable sharing of traced methods that are cafs. Methods definitions were always transformed as functions, even when they had arity 0. The reason was that an instance or class definition may not contain definitions for any other variables than the class variables. Such additional definitions are necessary to implement the sharing of cafs. Now the transformation distinguishes between the transformed definitions for the original variables and additional auxillary definitions. The latter are put just behind the instance/class definitions. The Mate trace file gets 1% smaller. Surprisingly runtime seems to slow down by 2-3%. M ./src/compiler98/DbgTrans.hs -86 +106

1 0

patch applied (hat): Completed tracing of named fields.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Apr 19 03:24:06 PDT 2001 olaf * Completed tracing of named fields. Now tracing of record construction and update is supported as well. However, the trace only shows the evaluation of the desugared program, that is, record construction is shown as data constructor application (possibly with `undefined' arguments) and record update is shown as use of the `case' construct. The trace file format needs to be extended to enable showing this operations properly. It would be nice if for all data constructors which have named fields these were always shown in the browser. Also: Corrected bug in records: x(C{x=1}{x=2}) evaluated to 1 instead of 2. Added missing type declaration of `readFile' to Prelude.T.hi. Handle the definition of `undefined' specially in the tracing transformation. `undefined' is treated as nullary function, not as caf. Thus it has a proper parent. Corrected combinators for nullary functions. They were too strict. M ./src/compiler98/DbgDataTrans.hs -4 +7 M ./src/compiler98/DbgTrans.hs -3 +33 M ./src/compiler98/Remove1_3.hs -51 +56 M ./src/compiler98/Type.hs -1 +10 M ./src/compiler98/TypeLib.hs -1 +4

1 0

patch applied (hat): nhc98 doesn' t like import of a named field via import Flags(sDbgTrans).
by Malcolm Wallace 10 Oct '06

10 Oct '06

Wed Apr 18 04:13:22 PDT 2001 malcolm * nhc98 doesn't like import of a named field via import Flags(sDbgTrans). Change to import Flags(Flags(..)). M ./src/compiler98/DbgTrans.hs -1 +1

1 0

patch applied (hat): Introduced even more Sats and resurrected field selectors.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Wed Apr 18 03:28:35 PDT 2001 olaf * Introduced even more Sats and resurrected field selectors. The application of a saturated application never had a Sat. Now it has. If a trusted application is recorded (because the function is suspected), then it has a Sat. There are now even more SatCs. Filesize grows by ~2%, runtime grows by ~2%. Corrected implementation of transformed field selectors. The definitions are derived at transformation time and subsequently transformed. Because nhc derives the type of imported selectors from data definitions, these types have to be patched by the transformation. Now application of a field selector to a value that does not contain the field is caught correctly in the transformed code. M ./src/compiler98/DbgDataTrans.hs -4 +2 M ./src/compiler98/DbgTrans.hs -64 +52 M ./src/compiler98/IExtract.hs -39 +127 M ./src/compiler98/Remove1_3.hs -2 +3

1 0

patch applied (hat): Reintroduced Sats
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Apr 12 10:08:16 PDT 2001 olaf * Reintroduced Sats Now in suspected code an application or Caf name only has no Sat, if its result is the same as the result of the parent. Note that in trusted code there are less Sats. A Sat is only constructed, if the unevaluated expression can be returned to suspected code (especially as argument of a constructor). Actually this is not quite true and needs to be corrected. There are now more SatCs than before. Filesize grows by 1%, speed change not noticeable. Undid part of previous implementation of field name selectors. They disturbed correct working of Cafs. So for now only selectors for data types that are defined in the same module will work. M ./src/compiler98/DbgTrans.hs -9 +18

1 0