- Cvs-nhc98 - Haskell.org

patch applied (hat): extended -v to check for null pointers in wrong context
by Malcolm Wallace 10 Oct '06

10 Oct '06

Fri Apr 6 06:49:42 PDT 2001 colin * extended -v to check for null pointers in wrong context added -n to print textual version of single node only extended -a and -n to show whether each node reachable or not with -r M ./src/hattools/hat-check.c -177 +233

1 0

patch applied (hat): Modified tracing transformation to create different programs for trusted and suspected modules.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Apr 5 12:13:29 PDT 2001 olaf * Modified tracing transformation to create different programs for trusted and suspected modules. In trusted modules no trace nodes for if, case, guards and lambdas are created. In trusted modules all trace nodes that are still created have the trace passed to the function or the hidden version of it as parent (hidden trace node is only created once in a trusted function). In trusted modules no trace nodes for the used functions are created. These have to be trusted functions (for the construction of the trace nodes for functions that are passed as arguments the module where the function is passed is responsible anyway). Some transformed trusted code is slightly simpler than suspected code (case, if, guards). Now have suspected and trusted versions of combinators fun and prim (tfun and tprim). The suspected ones always create a name for a function, the trusted ones only if the parent is not trusted. The trace node Root is now always trusted. This is needed to avoid names for trusted CAFs. This transformation profits from - enter and primEnter no longer requiring a table name: so with this transformation trusted table names are no longer constructed - Interact.hs being transformed as suspected: otherwise the output of a program using interact would directly point to Root. The new transformation leads to 10-20% smaller trace files and 10% faster runtime of traced programs. The main remaining unreachable parts of a trace seem to be: - data structures internal to trusted code - SAT's (all C?) for all trusted applications in non-project contexts - indirections in trusted code M ./src/compiler98/DbgId.hs -13 +21 M ./src/compiler98/DbgTrans.hs -548 +868 M ./src/compiler98/GcodeFix.hs -1 +6 M ./src/compiler98/Main.hs -2 +4 M ./src/compiler98/PrimCode.hs -13 +29 M ./src/old/runtime/fileformat.c -2 +2

1 0

patch applied (hat): Hat: primSR3() wrongly assumed ModInfo already in file.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Apr 5 07:57:50 PDT 2001 malcolm * Hat: primSR3() wrongly assumed ModInfo already in file. M ./src/old/runtime/fileformat.c +1

1 0

patch applied (hat): Hat: speed up cEnter, and remove its NmType argument.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Apr 5 07:41:15 PDT 2001 malcolm * Hat: speed up cEnter, and remove its NmType argument. Because cEnter is called on every reduction step, any improvement in its speed should go a long way. At the moment, it looks like we gain only about 1.5% by eliminating tests in C that are no longer useful. However, by eliminating its NmType argument, we hope that we can change some of the transformation/combinators, to produce much fewer NmType nodes in the file. M ./src/old/runtime/io.c -2 +8

1 0

patch applied (hat): Hat: On normal exit, use longjmp rather than explicit call of haskellEnd().
by Malcolm Wallace 10 Oct '06

10 Oct '06

Fri Mar 30 05:06:02 PST 2001 malcolm * Hat: On normal exit, use longjmp rather than explicit call of haskellEnd(). M ./src/old/runtime/io.c -2 +18

1 0

patch applied (hat): Extra comma in import list is not legal Haskell' 98.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Mar 29 13:21:23 PST 2001 malcolm * Extra comma in import list is not legal Haskell'98. M ./src/compiler98/DbgTrans.hs -2 +2

1 0

patch applied (hat): Add an FFI version of cHPutChar (for tracing).
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Mar 29 12:57:03 PST 2001 malcolm * Add an FFI version of cHPutChar (for tracing). M ./src/old/runtime/io.c +10

1 0

patch applied (hat): Improve infix notation in hat-stack output.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Mar 29 09:43:01 PST 2001 malcolm * Improve infix notation in hat-stack output. M ./src/hattools/hat-stack.c -4 +4 M ./src/hattools/utils.c -13 +73 M ./src/hattools/utils.h -1 +11

1 0

patch applied (hat): Minor tidy-up to columns in hat-check output.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Mar 29 09:42:31 PST 2001 malcolm * Minor tidy-up to columns in hat-check output. M ./src/hattools/hat-check.c -11 +11

1 0

patch applied (hat): Modified combinators and also slightly the transformation.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Wed Mar 28 10:33:56 PST 2001 olaf * Modified combinators and also slightly the transformation. Moved introduction of Sat's from functions side (fun?, prim?) to application side (ap?, rap?). - Removes previous bug that transformed program was too eager, e.g. in (True || h True) the application of h was evaluated. - Creation of Sat and App? trace nodes now next to each other. Integration of writing a Sat and an App? now straightforward. - Subsequent RHS-optimisation possible. Corrected bug in transformation of let. Expression after in is only in projective context if whole let is in projective context. This bug could lead to missing Hidden node in trusted code and too eager evaluation when the RHS-optimisation is used. Introduced separate combinator trif for if in projective context. The use of original tif could lead to too eager evaluation (never observed) and this would have become worse with the RHS-optimisation. RHS-optimisation: If an application (or if) is in projective context, then it doesn't need a Sat, because the application node only serves as a parent, but is never pointed to from an App? as an argument. So modified rap? and trif to not produce Sats. This halfs the number of SatCs and reduces runtime by ~10%. M ./src/compiler98/DbgId.hs -2 +4 M ./src/compiler98/DbgTrans.hs -4 +4

1 0