Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-01 01:54:15 -0800 (Fri, 01 Apr 2005)
Revision: 850
Log message:

      Explicit options to glob should override GLOB_OPTIONS, not the other way
      around.
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-04 22:56:23 -0700 (Mon, 04 Apr 2005)
Revision: 851
Log message:

      Bumped the version from 0.9.4.2 to 0.9.4.3.
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-05 00:22:43 -0700 (Tue, 05 Apr 2005)
Revision: 852
Log message:

      - Rearranged the usage message to have more important options on top.
      - Better error message when an invalid option is specified.
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-05 00:28:30 -0700 (Tue, 05 Apr 2005)
Revision: 853
Log message:

      Include the libdir in the --version output
      

Changes by: Nathaniel Gray (n8gray at cs.caltech.edu)
Date: 2005-04-06 15:15:14 -0700 (Wed, 06 Apr 2005)
Revision: 854
Log message:

      Use $(PATHSEP) instead of ':' in LaTeX rules.
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-08 15:17:48 -0700 (Fri, 08 Apr 2005)
Revision: 855
Log message:

      Adding a "dependencies" function that returns the set of dependencies.
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-08 15:31:30 -0700 (Fri, 08 Apr 2005)
Revision: 856
Log message:

      Documentation.
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-14 20:01:12 -0700 (Thu, 14 Apr 2005)
Revision: 857
Log message:

      Pervasives expects it to be called sequence-map, not sequence-foreach.
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-15 15:53:39 -0700 (Fri, 15 Apr 2005)
Revision: 858
Log message:

      Trivial: couple of simbols were defined twice.
      

Changes by: Nathaniel Gray (n8gray at cs.caltech.edu)
Date: 2005-04-16 04:14:33 -0700 (Sat, 16 Apr 2005)
Revision: 859
Log message:

      Did some cleanup of the fam_kqueue code.
      
      Also, omake doesn't care but FAMNextEvent is supposed to return 0 on success.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-16 15:04:43 -0700 (Sat, 16 Apr 2005)
Revision: 860
Log message:

      Preparing to fix the problem with regular-expression arguments
      
      http://cvs.cs.cornell.edu:12000/bugzilla/show_bug.cgi?id=315
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-17 09:11:34 -0700 (Sun, 17 Apr 2005)
Revision: 865
Log message:

      Added argument extraction.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-17 09:36:15 -0700 (Sun, 17 Apr 2005)
Revision: 866
Log message:

      This appears to be working, so after some clean up I'll commit it.
      TODO: add "earliest-match" arguments so that we can do searching
      in linear time.
      
      NOTE: This doesn't include StepIII in Aleksey's description (bug #315),
      because I don't understand the need for it.
      
      The consequences are that the semantics are last-match.  For example:
      
         match $'fooXbooXmooXgoo'
         case $'\((\(.*\)X)*\)'
            println($(string $*))
      
      prints
         fooXbooXmooX moo
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-17 11:29:03 -0700 (Sun, 17 Apr 2005)
Revision: 868
Log message:

      This is the linear-time version of the search.  This changes
      the semantics of the search function.  I think the old one, if
      it failed, would return the text to the eof.  This one just
      fails, so I'll have to see if these other files need to be
      updated.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-17 11:51:18 -0700 (Sun, 17 Apr 2005)
Revision: 869
Log message:

      Fixed some typos so searching actually works.  It is currently
      longest-match, so this will have to be fixed.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-18 12:43:35 -0700 (Mon, 18 Apr 2005)
Revision: 871
Log message:

      Now that we have separate arg registers, we can be more aggressive
      about computing the epsilon closure through argument calculations.
      This is a pre-commit before I delete some of the code.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-18 12:56:09 -0700 (Mon, 18 Apr 2005)
Revision: 872
Log message:

      Deleted the unnecessary code, and simplified arg representation a bit.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-18 15:32:25 -0700 (Mon, 18 Apr 2005)
Revision: 873
Log message:

      The lexer is pretty much finished.  I want to try it in MetaPRL next.
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-18 18:07:52 -0700 (Mon, 18 Apr 2005)
Revision: 874
Log message:

      Documentation update.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-19 12:03:29 -0700 (Tue, 19 Apr 2005)
Revision: 875
Log message:

      Added Lexer.searchto function, for matching against delimiters.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-19 12:18:23 -0700 (Tue, 19 Apr 2005)
Revision: 877
Log message:

      Merged the lexer_args_correction branch.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-19 12:18:56 -0700 (Tue, 19 Apr 2005)
Revision: 878
Log message:

      Bumped the version number.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-19 12:42:49 -0700 (Tue, 19 Apr 2005)
Revision: 879
Log message:

      One problem was that we were not using readline on Linux when
      compiled with omake (the Makefile.in was ok).
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-19 12:46:25 -0700 (Tue, 19 Apr 2005)
Revision: 880
Log message:

      Instead of raising End_of_file, set the return length to 0, when
      not using GNU readline in readline.c.  This will prevent the
      lexer from going crazy with indentation warnings.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-19 13:15:45 -0700 (Tue, 19 Apr 2005)
Revision: 881
Log message:

      Added better error handling during the interactive shell.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-19 13:21:24 -0700 (Tue, 19 Apr 2005)
Revision: 882
Log message:

      Fix white space in lexer.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-19 14:34:17 -0700 (Tue, 19 Apr 2005)
Revision: 883
Log message:

      Added EXIT_ON_SHELL_ERROR to raise an exception when a shell
      command returns an error in an expression evaluation (bug #397).
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-19 14:46:18 -0700 (Tue, 19 Apr 2005)
Revision: 884
Log message:

      osh should exit when a shell command fails and EXIT_ON_SHELL_ERROR is set.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-19 15:47:13 -0700 (Tue, 19 Apr 2005)
Revision: 885
Log message:

      Due to Aleksey's (implicit) suggestion, I've redefined the variables:
      
         ABORT_ON_COMMAND_ERROR: if defined and true, raises an
            exception whenever a shell command terminates abnormally.
         EXIT_ON_UNCAUGHT_EXCEPTION: if defined and true, causes
            osh to exit abnormally whenver an exception is not
            caught.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-19 18:31:29 -0700 (Tue, 19 Apr 2005)
Revision: 886
Log message:

      Added initial code to compute the digest of a value.  This code
      works the naive way, by converting the value to string form,
      then computing the MD5 digest.  An alternate approach would be
      to convert the value to a simplified form (a form that does not
      include environments), then use Pervasives.compare.  The
      construction phase is a bit cheaper, but it will blow up space
      usage, so it is a tradeoff.
      
      This is the initial step toward fixing bug# 431.
      

Changes by: ( at unknown.email)
Date: 2005-04-19 18:31:29 -0700 (Tue, 19 Apr 2005)
Revision: 887
Log message:

      This commit was manufactured by cvs2svn to create branch
      'value_dependencies'.

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-19 19:21:46 -0700 (Tue, 19 Apr 2005)
Revision: 888
Log message:

      It is clear that I should work on a branch while adding value-dependencies.
      
      This is a non-compiling version that adds value-dependencies to the
      rule info, but doesn't yet propagate it all the way to the build.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-21 13:20:03 -0700 (Thu, 21 Apr 2005)
Revision: 889
Log message:

      Fixed a subtle error regarding shortest match.
      When a final state is reached, any NFA component of the DFA
      state should be removed if either of the following hold:
         1. the state is part of the search prefix
         2. the state came from a state in the search prefix, unless
            it is the final state being considered.
      Previously, we did part 1, but not part 2.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-21 14:04:40 -0700 (Thu, 21 Apr 2005)
Revision: 890
Log message:

      Propagate value information through Omake_build.  This still needs to
      be evaluated and added to the cache.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-21 17:27:12 -0700 (Thu, 21 Apr 2005)
Revision: 891
Log message:

      Partially propagated value digests through the build.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-22 16:29:18 -0700 (Fri, 22 Apr 2005)
Revision: 892
Log message:

      Simplifying the value dependencies a bit.  In the build, it is simpler
      to treat the value dependencies just as a special command line that
      acts as a no-op, but changes to the values are treated as if the
      commands changed.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-23 09:43:30 -0700 (Sat, 23 Apr 2005)
Revision: 893
Log message:

      Split the definitions out of the cache.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-23 12:18:49 -0700 (Sat, 23 Apr 2005)
Revision: 894
Log message:

      Propagated the value dependencies all the way to the cache.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-23 12:49:20 -0700 (Sat, 23 Apr 2005)
Revision: 895
Log message:

      Correct :values: option.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-23 12:53:50 -0700 (Sat, 23 Apr 2005)
Revision: 896
Log message:

      Rename omake_value_digest -> omake_command_digest
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-23 14:05:55 -0700 (Sat, 23 Apr 2005)
Revision: 897
Log message:

      Add free variables of commend expressions as value dependencies.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-24 12:36:10 -0700 (Sun, 24 Apr 2005)
Revision: 898
Log message:

      Reasonably fast, but ad-hoc digest function.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-25 18:47:14 -0700 (Mon, 25 Apr 2005)
Revision: 899
Log message:

      Minor change to ad-hoc digest.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-26 12:33:48 -0700 (Tue, 26 Apr 2005)
Revision: 900
Log message:

      Values were being dropped during rule evaluation.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-26 12:46:55 -0700 (Tue, 26 Apr 2005)
Revision: 901
Log message:

      Added value-dependencies.
      
      Value dependencies are specified using the :value: option in rules.
      
         a: b c :value: $(X)
            ...
      
      This rule specifies that "a" should be recompiled if the value of $(X)
      changes (X does not have to be a filename).  This is intended to allow
      greater control over dependencies.
      
      In addition, it can be instead of other kinds of dependencies.
      For example, the following rule:
      
         a: b :exists: c
            commands
      
      is the same as
      
         a: b :value: $(target-exists c)
      
      
      Notes:
      
         -- The values are arbitrary (they are not limited to variables)
         -- The values are evaluated at rule expansion time, so variables
            like $@, $^, etc are legal.
      
      One other significant difference is that the rule cache now uses a
      digest of the rule commands text, not the text itself.  This has an
      impact on initial omake speed (I am not sure how significant it is)
      but the cache is smaller.  Also, "section eval" should now be
      handled correctly.
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-26 13:12:21 -0700 (Tue, 26 Apr 2005)
Revision: 902
Log message:

      Added a section documenting :value: dependencies (based on Jason's commit
      message).
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-26 14:44:03 -0700 (Tue, 26 Apr 2005)
Revision: 903
Log message:

      Fixed the buffer problem in .SCANNER rules.  Actually, there was even
      a comment "JYH: I don't think checking for stdout and stderr is
      necessary."  Well, it is.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-26 15:04:51 -0700 (Tue, 26 Apr 2005)
Revision: 904
Log message:

      A silly bug.  For $(fun ...), the body was evaluated, the result
      discarded, and the original body returned as the result:!
      The addresses #452.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-26 15:16:45 -0700 (Tue, 26 Apr 2005)
Revision: 905
Log message:

      For some reason, shell expressions were being evaluated lazily.
      This seems like a bad idea.  Changed evaluation to be eager by
      default, addressing bug #454.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-26 15:37:40 -0700 (Tue, 26 Apr 2005)
Revision: 906
Log message:

      Strip sequence objects in values_of_value.  This addresses bug #449.
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-26 16:05:15 -0700 (Tue, 26 Apr 2005)
Revision: 907
Log message:

      Added a .SCANNER rule for [La]TeX files.
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-26 16:22:10 -0700 (Tue, 26 Apr 2005)
Revision: 908
Log message:

      Simplified builtin-tex-deps a bit.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-26 17:54:11 -0700 (Tue, 26 Apr 2005)
Revision: 909
Log message:

      Change the scoping for unknown variables to the senv_mode, so
      variables in an object will default to ScopeProtected.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-27 11:26:32 -0700 (Wed, 27 Apr 2005)
Revision: 910
Log message:

      Added the syntax
      
         $'key string' = value
      
      for defining maps.  The $"..." also works, so you can add variable
      expansions if you like.
      
         KEY = key string
         $"$(KEY) 1" = value
      
      The model for maps is also changed.  Every object has a "property
      table," which is just a map of this form.  You can attach properties
      to any object.
      
         AnnotatedOne. =
            extends $(object $(int 1))
            $'description' = This is the number one
      
      The "Map" class just makes it easier to access the properties.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-29 12:44:42 -0700 (Fri, 29 Apr 2005)
Revision: 912
Log message:

      Made some changes to the Map class.
      
      The keys are now "simple" values, not just strings.  These
      include integers, floats, strings, arrays of simple values,
      files, and directories.
      
      Only the Map class has the map functions defined.  If you
      want to have, say, a File that also includes a Map,
      create a subclass that extends both File and Map.
      
      The literal keys now have the form $|key...|.  This works
      both for definitions and uses.  The usual modifiers are
      allowed $,|key| and $`|key|.
      
         X. =
            extends $(Map)
            $|key 1| = 1
            $|key 2| = $|key 1| boo
            $|key 2| += foo
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-29 13:13:12 -0700 (Fri, 29 Apr 2005)
Revision: 913
Log message:

      Do the right thing with PWD in builtin-tex-deps.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-29 14:06:30 -0700 (Fri, 29 Apr 2005)
Revision: 914
Log message:

      Added a 2-argument version of cd.  To evaluate $(cd dir, e),
      change to the directory "dir", then exvaluate expression "e"
      and return the result.  The expression "e" is evaluated lazily.
      The current directory is not changed otherwise.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-29 14:27:17 -0700 (Fri, 29 Apr 2005)
Revision: 915
Log message:

      $(dirof ...) should be smarter about computing the directory.
      For $(dirof x), if "x" already a directory, return it.
      This address bug #379.
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-29 14:41:17 -0700 (Fri, 29 Apr 2005)
Revision: 916
Log message:

      Documentation fix: ln is not a builtin command.
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-29 15:03:01 -0700 (Fri, 29 Apr 2005)
Revision: 917
Log message:

      Use the new 2-arg version of cd. I bumped the version number to 0.9.4.6 to
      avoid accidentally using new OMakeroot.src with older omake w/o 2-arg cd.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-29 17:19:33 -0700 (Fri, 29 Apr 2005)
Revision: 918
Log message:

      Fixed a bug introduced during value dependencies, where
      Omake_build.is_leaf_node was not working correctly because
      it got confused by a non-empty command list.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-29 17:25:28 -0700 (Fri, 29 Apr 2005)
Revision: 919
Log message:

      Change is-leaf-node so that targets with value-dependencies are
      non-leaves.
      
      Also, in the previous commit, changed += so that it does not append,
      nor separate, "empty" values (as defined in Omake_env.is_empty_value).
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-30 15:24:33 -0700 (Sat, 30 Apr 2005)
Revision: 920
Log message:

      When splitting a sequence into a list of values along the space boundaries, do
      not stringify primitive values (such as files and directories). This partially
      solves bug #443.
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-30 16:52:34 -0700 (Sat, 30 Apr 2005)
Revision: 921
Log message:

      Working on outstanding issues for Win32.
      
      This is a fix to be more careful about spaces in filenames,
      so that placing ocaml in
      
         C:\Program Files\Objective Caml
      
      actually works.  This adds the $(quote-argv ...) function that
      quotes an array in a way that the command-line parser can recover
      the argv string.  This allows include directories with spaces in
      a -pp option.
      

Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2005-04-30 17:04:34 -0700 (Sat, 30 Apr 2005)
Revision: 922
Log message:

      Be less agressive in turning the sequences into arrays on "+="
      (bug 443, bug 465).
      

Changes by: Jason Hickey (jyh at cs.caltech.edu)
Date: 2005-04-30 18:04:48 -0700 (Sat, 30 Apr 2005)
Revision: 923
Log message:

      Make scanner dependencies soft recursive.  That is, the results of
      a scan become soft dependencies for the next scan.
      
      This will work fairly well for C, addressing bug #423, but it
      is too aggressive for OCaml.  I'll try to address that next.