Add -i option, append .awk if initial search fails, and allow -f repeats.

1 files changed, 113 insertions, 88 deletions

diff --git a/TODO.xgawk b/TODO.xgawk
index 1917c73d..04b82b5e 100644
--- a/TODO.xgawk
+++ b/TODO.xgawk
@@ -1,86 +1,37 @@
 To-do list for xgawk enhancements:
 
+- Finish implementing new interface using gawkapi.h
 
-Done:
-
-- Add AWKLIBPATH with default pointing to ${libexecdir}/$PACKAGE/$VERSION
-
-- Change default shared library extension from ".so" to ".$shlibext"
-
-- Patch build infrastructure so that the current code in the
-  extension subdirectory gets built and installed into the default $AWKLIBPATH
-  location.
-
-- Implement @load
-
-- Patch ERRNO handling to create a simple API for use by extensions:
-   extern void update_ERRNO_int(int)
-   enum errno_translate { TRANSLATE, DONT_TRANSLATE };
-   extern void update_ERRNO_string(const char *string, enum errno_translate);
-   extern void unset_ERRNO(void);
-
-- Add valgrind-noleak target.
-
-- Fix minor bug in fork extension, and add wait function.
-
-- Patch filefuncs extension to read symbolic links more robustly.
-
-- Add shared library tests.
-
-- Delete extension/xreadlink.[ch]
-
-
-To do (not necessarily in this order):
-
-- Add a -i (--include) option.
-
-- The -f flag should not eliminate duplicates.
-
-- Enable default ".awk" search in io.c:find_source().  The simple change
-  is to add this code inline in io.c:
-      #ifndef DEFAULT_FILETYPE
-      #define DEFAULT_FILETYPE ".awk"
-      #endif
-
-- Implement new interface using gawkapi.h
+- Consider behavior when -i and -f include the same file.  Currently,
+  -f always loads the code, regardless of whether it has already been loaded
+  using -i or -f.  By contrast, -i will ignore the file if has already been
+  loaded with either -i or -f.  Thus, if you say "gawk -f lib.awk -i lib.awk",
+  it will not be included the second time.  But if you flip the order
+  and say "gawk -i lib.awk -f lib.awk", the file will be loaded twice.
+  Is this the most sensible behavior?
 
 - Add time extension to the gawk distro. This defines sleep and gettimeofday.
-  Rename existing gettimeofday by adding some underscores.
+  Rename existing gettimeofday by adding some underscores.  Awaiting
+  confirmation of copyright assignment from FSF...
 
-- Enhance extension/fork.c waitpid to allow the caller to specify the options.
-  And add an optional array argument to wait and waitpid in which to return
-  exit status information.
+- Running "make install" should install gawkapi.h in /usr/include.
 
-- Maybe add more shared library tests.
+- Decide how to transition from the old extension API to the new one.
+  When will the old approach be abandoned?
 
-- Fix extension/rwarray.c.  It does not currently compile due to changes
-  in the NODE structure relating to array support.  The MPFR changes will
-  also make this more complicated.
-
-- Figure out how to support xgawk on platforms such as Cygwin where a DLL
-  cannot be linked with unresolved references.  There are currently 3
-  possible solutions:
-    1. Restructure gawk as a stub calling into a shared library.
-    2. Move a subset of gawk interfaces into a shared library that can be
-       called by extensions.
-    3. Change the interface between gawk and extensions so that gawk will
-       pass a pointer to a structure into dlload that contains the addresses
-       of all variables and functions to which the extension may need access.
+- Eliminate libtool from the top-level configure.ac.  Create a separate
+  configure.ac in the extensions subdirectory, and hide all the libtool
+  stuff in there.  I think AC_CONFIG_SUBDIRS may do this.  Should building
+  the extensions be optional (e.g. by adding a --without-extensions option
+  to the top-level configure)?  If not, I do not understand what is
+  accomplished by using a separate configure script for the extensions...
 
-- Fix lint complaints about shared library functions being called without
-  having been defined.  For example, try:
-     gawk --lint -lordchr 'BEGIN {print chr(65)}'
-     gawk: warning: function `chr' called but never defined
-     A
-  In ext.c, make_builtin needs to call awkgram.y:func_use.  If done naively,
-  I think this would result in complaints about shared library functions
-  defined but not used.  So there should probably be an enhancement to func_use
-  and ftable to indicate if it's a shared library function.
+- Develop a libgawk shared library for use by extensions.  Should this
+  be hosted in a separate project?
 
-- Develop a libgawk shared library for use by extensions.  In particular,
-  a few existing extensions use a hash API for mapping string handles to
-  structures.  In xgawk, we had this API inside array.c, but it probably
-  belongs in a separate libgawk shared library:
+  A few existing extensions use a hash API for mapping string
+  handles to structures.  In xgawk, we had this API inside array.c, but it
+  probably belongs in a separate libgawk shared library:
 
   typedef struct _strhash strhash;
   extern strhash *strhash_create P((size_t min_table_size));
@@ -96,19 +47,28 @@ To do (not necessarily in this order):
 			       strhash_delete_func, void *opaque));
   extern void strhash_destroy P((strhash *, strhash_delete_func, void *opaque));
 
-- Running "make install" should install the new libgawk shared library
-  as well as header files needed to build extensions under /usr/include/gawk.
-  The extensions include "awk.h", and that pulls in the following headers
-  (according to gcc -M) :
-      awk.h config.h custom.h gettext.h mbsupport.h protos.h getopt.h \
-      regex.h dfa.h
-  Most likely, most of this is not required.  Arnold has suggested 
-  creating a smaller header to define the public interface for use by shared
-  libraries.  One could imagine having "awk-ext.h" that is included by "awk.h".
-
-
-Separate projects for major standalone extensions.  Where should these
-be hosted?
+- Review open hook implementation.  Arnold's comments on this:
+   I think the code flow in io.c needs reviewing. It's not
+   clear to me under what circumstances open hooks (a) are called, or
+   (b) should be called.  Only the XML extension uses them now, but I
+   think it'd be a nice mechanism to generalize if possible, and to
+   document.
+   
+   E.g., I can easily envision an open hook to turn directories into records
+   of the form
+  
+         type link-count mode owner group atime mtime ctime name
+   
+   I could also envision an open hook to provide an interface to libiconv
+   (somehow).  More discussion / explanation of the vision behind this
+   would be welcome.
+
+- Can the IOBUF internals be removed from gawkapi.h?  I think this may be
+  possible if we revise the open hook implementation.
+
+
+Separate projects for major standalone extensions.  We need to set up
+hosting for these projects:
 
 - XML
 
@@ -116,10 +76,33 @@ be hosted?
 
 - GD
 
-- MPFR.  Is this still useful if MPFR support will be integrated into gawk?
+- MPFR.  This is probably not useful now that MPFR support has been
+  integrated into gawk.  Are there any users who need this extension?
+
+
+Low priority:
 
+- Fix extension/rwarray.c.  It does not currently compile due to changes
+  in the NODE structure relating to array support.  The MPFR changes will
+  also make this more complicated.  John is best equipped to solve this
+  problem.
+
+- Enhance extension/fork.c waitpid to allow the caller to specify the options.
+  And add an optional array argument to wait and waitpid in which to return
+  exit status information.
 
-Possible changes requiring (further) discussion:
+- Fix lint complaints about shared library functions being called without
+  having been defined.  For example, try:
+     gawk --lint -lordchr 'BEGIN {print chr(65)}'
+     gawk: warning: function `chr' called but never defined
+     A
+  In ext.c, make_builtin needs to call awkgram.y:func_use.  If done naively,
+  I think this would result in complaints about shared library functions
+  defined but not used.  So there should probably be an enhancement to func_use
+  and ftable to indicate if it's a shared library function.
+
+
+Possible future changes requiring (further) discussion:
 
 - Change from dlopen to using the libltdl library (i.e. lt_dlopen).
   This may support more platforms.
@@ -139,6 +122,48 @@ Possible changes requiring (further) discussion:
   etc, I'd like to keep things simple.  But how we design this is going
   to be very important.
 
+
+Unlikely:
+
 - Include a sample rpm spec file in a new packaging subdirectory.
 
 - Patch lexer for @include and @load to make quotes optional.
+
+
+Done:
+
+- Add AWKLIBPATH with default pointing to ${libexecdir}/$PACKAGE/$VERSION
+
+- Change default shared library extension from ".so" to ".$shlibext"
+
+- Patch build infrastructure so that the current code in the
+  extension subdirectory gets built and installed into the default $AWKLIBPATH
+  location.
+
+- Implement @load
+
+- Patch ERRNO handling to create a simple API for use by extensions:
+   extern void update_ERRNO_int(int)
+   enum errno_translate { TRANSLATE, DONT_TRANSLATE };
+   extern void update_ERRNO_string(const char *string, enum errno_translate);
+   extern void unset_ERRNO(void);
+
+- Add valgrind-noleak target.
+
+- Fix minor bug in fork extension, and add wait function.
+
+- Patch filefuncs extension to read symbolic links more robustly.
+
+- Add shared library tests.
+
+- Delete extension/xreadlink.[ch]
+
+- Add a -i (--include) option.
+
+- Enable default ".awk" search in io.c:find_source().  The simple change
+  is to add this code inline in io.c:
+      #ifndef DEFAULT_FILETYPE
+      #define DEFAULT_FILETYPE ".awk"
+      #endif
+
+- The -f flag should not eliminate duplicates.