- Cvs-nhc98 - Haskell.org

patch applied (hat): Add Phil' s time profiler into the standard code base.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu May 3 10:17:04 PDT 2001 malcolm * Add Phil's time profiler into the standard code base. * quite a few additions to the runtime system * some mangling of ld-lines required in scripts * in compiler, introduce an extra field in InfoName constructor to identify local function definitions (i.e. non-globals) for correct aggregation of time costs. M ./src/compiler98/Case.hs -3 +3 M ./src/compiler98/CaseLib.hs -1 +1 M ./src/compiler98/FSLib.hs -1 +1 M ./src/compiler98/GcodeFix.hs -4 +10 M ./src/compiler98/GcodeLow.hs -1 +4 M ./src/compiler98/Info.hs -8 +9 M ./src/compiler98/IntState.hs -1 +1 M ./src/compiler98/Lift.hs -2 +2 M ./src/compiler98/PosAtom.hs -2 +2 M ./src/compiler98/RenameLib.hs -1 +1 M ./src/compiler98/ReportImports.hs -1 +1

1 0

patch applied (hat): Recent changes to DbgTrans.hs require more stack space when building
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu May 3 09:41:53 PDT 2001 malcolm * Recent changes to DbgTrans.hs require more stack space when building the compiler. M ./src/compiler98/Makefile -2 +2

1 0

patch applied (hat): Add Thorsten' s hat tools to the larger set.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Wed May 2 07:01:28 PDT 2001 malcolm * Add Thorsten's hat tools to the larger set. * hat-observe for HOOD-like observation of functions * hat-detect for Freja-like algorithmic debugging * hat-interact for examining hat-check listings Build rules for these are incorporated into the Makefile. Also, move some of 'browsercomms.c' into a new main prog 'hat-connect.c', just for consistency with the other tools. A ./src/hattools/Expressions.c A ./src/hattools/Expressions.h A ./src/hattools/FunTable.c A ./src/hattools/FunTable.h M ./src/hattools/Makefile -25 +40 M ./src/hattools/browsercomms.c -47 +2 A ./src/hattools/hashtable.c A ./src/hattools/hashtable.h A ./src/hattools/hat-connect.c A ./src/hattools/hat-detect.c A ./src/hattools/hat-interact.c A ./src/hattools/hat-observe.c A ./src/hattools/hatfile.h A ./src/hattools/hatfileops.c A ./src/hattools/hatfileops.h

1 0

patch applied (hat): Improve runtime code from -O2 to -O3 (options to gcc).
by Malcolm Wallace 10 Oct '06

10 Oct '06

Wed May 2 05:49:32 PDT 2001 malcolm * Improve runtime code from -O2 to -O3 (options to gcc). M ./Makefile.inc -1 +1

1 0

patch applied (hat): Introduce Sat for outermost application on rhs of pattern binding.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Fri Apr 27 02:53:55 PDT 2001 olaf * Introduce Sat for outermost application on rhs of pattern binding. The outermost application on a rhs usually doesn't have a Sat, because the result equals the result of the parent, which is the lhs. However, for pattern bindings the lhs is not the parent. So hat-observe needs the Sat in that case. If the rhs of a pattern binding is a single lambda-bound variable, we no longer create an indirection. However, this indirection seems pretty pointless anyway, the parent of the variable value is probably more interesting. Increases Mate trace file by 0.02%. M ./src/compiler98/DbgTrans.hs -17 +15

1 0

patch applied (hat): Add variable names to top-level trusted pattern bindings.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Apr 26 10:28:19 PDT 2001 olaf * Add variable names to top-level trusted pattern bindings. Similar to top-level trusted cafs. In practise this probably hardly occurs. However, this variable names are only shown if the variables are not evaluated. If they are, they are no longer reachable in the trace. That may mean, that there is only root as printable representation of the result. However, it is unclear where the variable names should be added to the trace. M ./src/compiler98/DbgTrans.hs -1 +48

1 0

patch applied (hat): Improved trace construction for suspected pattern bindings.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Thu Apr 26 08:31:46 PDT 2001 olaf * Improved trace construction for suspected pattern bindings. Introduced an indirection in transformation of suspected pattern bindings, so that the whole rhs is reachable in hat-trail. However, the connection to the names of the variables are lost, when they are evaluated. There seems to be no natural point where to add them to the trace. Increases Mate trace file by 1%. Corrected bug in hat-trail which made it impossible to see parents of indirections in many cases. However, hat-trail still does not deal with indirections completely correctly. Pointing at an expression surrounded by an indirection highlights the parent of the expression in yellow, not the parent the indirection points to (and which is shown if asked for). First thought that hat-connect doesn't deal with indirections correctly, but it is in fact correct. M ./src/compiler98/DbgTrans.hs -4 +7 M ./src/hattools/browsercomms.c -2 +2 M ./src/old/trail/DbgPanel.java -13 +13

1 0

patch applied (hat): Top-level trusted cafs get names.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Wed Apr 25 10:25:33 PDT 2001 olaf * Top-level trusted cafs get names. Top-level trusted cafs (such as concat and unlines) now have a name node in the trace, so that this name can be shown. Note that only top-level trusted cafs can be called from suspected code (that local trusted cafs might be returned to suspected code is irrelevant; because this code is trusted, it should not be shown (just a hidden tag)). Something similar should probably be done for top-level trusted pattern bindings. However, these need some redesign in other respects also. Also changed transformation so that no source reference table entries for empty source references (noPos) are created. M ./src/compiler98/DbgId.hs -3 +3 M ./src/compiler98/DbgTrans.hs -14 +67

1 0

patch applied (hat): Fix modinfo pointer-alignment in obsolete section of code.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Wed Apr 25 06:13:17 PDT 2001 malcolm * Fix modinfo pointer-alignment in obsolete section of code. (Committed just in case we ever decide to resurrect the old stuff.) M ./src/old/runtime/browsercomms.c -3 +3

1 0

patch applied (hat): Avoid a coredump if modinfo does not exist.
by Malcolm Wallace 10 Oct '06

10 Oct '06

Wed Apr 25 06:10:55 PDT 2001 malcolm * Avoid a coredump if modinfo does not exist. M ./src/old/runtime/fileformat.c -1 +4

1 0