Patchwork gengtype improvements for plugins, thirdround! patch 4/7 [filerules]

login
register
mail settings
Submitter Basile Starynkevitch
Date Sept. 21, 2010, 7:16 p.m.
Message ID <20100921211628.969ed469.basile@starynkevitch.net>
Download mbox | patch
Permalink /patch/65373/
State New
Headers show

Comments

Basile Starynkevitch - Sept. 21, 2010, 7:16 p.m.
Hello All,

[join work by Basile Starynkevitch & Jeremie Salvucci]

References: 
  http://gcc.gnu.org/ml/gcc-patches/2010-08/msg02060.html
  http://gcc.gnu.org/ml/gcc-patches/2010-09/msg00616.html
  http://gcc.gnu.org/ml/gcc-patches/2010-09/msg00663.html
  http://gcc.gnu.org/ml/gcc-patches/2010-08/msg02063.html
  http://gcc.gnu.org/ml/gcc-patches/2010-08/msg02065.html
  http://gcc.gnu.org/ml/gcc-patches/2010-09/msg01709.html

This is a slightly improved version of the patch
http://gcc.gnu.org/ml/gcc-patches/2010-08/msg02065.html with Laurynas
comments from http://gcc.gnu.org/ml/gcc-patches/2010-09/msg00419.html
adressed.  As I told before, it improves the core
get_output_file_with_visibility function. This version (relative to
http://gcc.gnu.org/ml/gcc-patches/2010-08/msg02065.html I did sent
before) improved comments and function names, following Laurynas
advices; I also avoided calling regexec twice needlessly. I added a
lot of explanatory comments.  It is made much more modular and less ad
hoc by a file rule machinery. We now use regular expressions to match
the input file name and compute the associated output name & for name.

A detailed comment is available in the patch.

The patch relative to http://gcc.gnu.org/ml/gcc-patches/2010-09/msg01709.html is obtained
with

diff -u -p  $(svn stat . |awk '/[AM]/{print $2}') \
     --from-file ../thirdround_03_inputfile/ > \
     $HOME/Gengtype_Thirdround/patch4_filesrules-relto03.diff

########## gcc/ChangeLog entry ###########"
2010-09-21  Jeremie Salvucci  <jeremie.salvucci@free.fr>
	    Basile Starynkevitch  <basile@starynkevitch.net>

	* gengtype.c: Include xregex.h and obstack.h
	Added comments about role of get_output_file_with_visibility and
	our regexpr machinery.
	(frul_actionrout_t, struct file_rule_st): New.
	(hader_dot_h_frul, source_dot_c_frul): New functions.
	(NULL_REGEX,  NULL_FRULACT): New.
	(files_rules): New.
	(matching_file_name_substitute): New function.
	(get_output_file_with_visibility): Updated comments and rewritten
	to use the new files_rules machinery.
#######################


For convenience, I am also attaching the cumulated patches against trunk 
Ok for trunk?

Cheers.
Laurynas Biveinis - Sept. 22, 2010, 2:32 a.m.
+/* Special file rules action for handling *.c source files using
+ * get_file_gtfilename to compute their output_name and
+ * get_file_basename to compute their for_name.  The output_name is
+ * gt-<LANG>-<BASE>.h for language specific source files, and
+ * gt-<BASE>.h for common source files.  */

Here and everywhere else, please drop the leading columns of '*'.
Otherwise, the patch looks OK to me. (I was hoping for net code size
savings here, but it looks like regexps only balloon it. Anyway, I
still like the patch for its data-driveness).

2010/9/21 Basile Starynkevitch <basile@starynkevitch.net>:
> 2010-09-21  Jeremie Salvucci  <jeremie.salvucci@free.fr>
>            Basile Starynkevitch  <basile@starynkevitch.net>
>
>        * gengtype.c: Include xregex.h and obstack.h
>        Added comments about role of get_output_file_with_visibility and
>        our regexpr machinery.
>        (frul_actionrout_t, struct file_rule_st): New.
>        (hader_dot_h_frul, source_dot_c_frul): New functions.
>        (NULL_REGEX,  NULL_FRULACT): New.
>        (files_rules): New.
>        (matching_file_name_substitute): New function.
>        (get_output_file_with_visibility): Updated comments and rewritten
>        to use the new files_rules machinery.

Patch

--- ../thirdround_03_inputfile//gengtype.c	2010-09-21 11:02:01.000000000 +0200
+++ gcc/gengtype.c	2010-09-21 13:53:03.000000000 +0200
@@ -25,6 +25,8 @@ 
 #include "double-int.h"
 #include "version.h"    /* for version_string & pkgversion_string.  */
 #include "hashtab.h"
+#include "xregex.h"
+#include "obstack.h"
 #include "gengtype.h"
 
 /* Data types, macros, etc. used only in this file.  */
@@ -854,7 +856,7 @@  create_field_all (pair_p next, type_p ty
   field->type = type;
   field->name = name;
   field->opt = opt;
-  field->line.file = inpf;
+  field->line.file = CONST_CAST (input_file*, inpf);
   field->line.line = line;
   return field;
 }
@@ -1727,18 +1729,259 @@  get_file_gtfilename (const input_file *i
   return result;
 }
 
-/* An output file, suitable for definitions, that can see declarations
-   made in INPUT_FILE and is linked into every language that uses
-   INPUT_FILE.  */
+
+/** Each input_file has its associated output file outf_p.  The
+    association is computed by the function
+    get_output_file_with_visibility.  The associated file is cached
+    inside input_file in its inpoutf field, so is really computed only
+    once.  Associated output file paths (i.e. output_name-s) are
+    computed by a rule based regexp machinery, using the files_rules
+    array of struct file_rule_st.  A for_name is also computed, giving
+    the source file name for which the output_file is generated; it is
+    often the last component of the input_file path.  */
+
+
+/***
+ Regexpr machinery to compute the output_name and for_name-s of each
+ input_file.  We have a sequence of file rules which gives the POSIX
+ extended regular expression to match an input file path, and two
+ transformed strings for the corresponding output_name and the
+ corresponding for_name.  The transformed string contain dollars: $0
+ is replaced by the entire match, $1 is replaced by the substring
+ matching the first parenthesis in the regexp, etc.  And $$ is replaced
+ by a single verbatim dollar.  The rule order is important.  The
+ general case is last, and the particular cases should come before.
+ An action routine can, when needed, update the out_name & for_name
+ and/or return the appropriate output file.  It is invoked only when a
+ rule is triggered.  When a rule is triggered, the output_name and
+ for_name are computed using their transform string in while $$, $0,
+ $1, ... are suitably replaced.  If there is an action, it is called.
+ In some few cases, the action can directly return the outf_p, but
+ usually it just updates the output_name and for_name so should free
+ them before replacing them.  The get_output_file_with_visibility
+ function creates an outf_p only once per each output_name, so it
+ scans the output_files list for previously seen output file names.
+ */
+
+/* Signature of actions in file rules.  */
+typedef outf_p (frul_actionrout_t) (input_file*, char**, char**);
+
+
+struct file_rule_st {
+  const char* frul_srcexpr;	/* Source string for regexp.  */
+  int frul_rflags;		/* Flags passed to regcomp, usually
+				 * REG_EXTENDED.  */
+  regex_t* frul_re;		/* Compiled regular expression
+				   obtained by regcomp.  */
+  const char* frul_tr_out;	/* Transform string for making the
+				 * output_name, with $1 ... $9 for
+				 * subpatterns and $0 for the whole
+				 * matched filename.  */
+  const char* frul_tr_for;	/* Tranform string for making the
+				   for_name.  */
+  frul_actionrout_t* frul_action; /* The action, if non null, is
+				   * called once the rule matches, on
+				   * the transformed out_name &
+				   * for_name.  It could change them
+				   * and/or give the output file.  */
+};
+
+/* File rule action handling *.h files.  */
+static outf_p header_dot_h_frul (input_file*, char**, char**);
+
+/* File rule action handling *.c files.  */
+static outf_p source_dot_c_frul (input_file*, char**, char**);
+
+#define NULL_REGEX (regex_t*)0
+#define NULL_FRULACT (frul_actionrout_t*)0
+
+/* The array of our rules governing file name generation.  Rules order
+   matters!  Change these rules with extreme care!  */
+
+struct file_rule_st files_rules[] = {
+  /* the c-family/ source directory is special.  */
+  { "^(([^/]*/)*)c-family/([[:alnum:]_-]*)\\.c$",
+    REG_EXTENDED, NULL_REGEX,
+    "gt-c-family-$3.h", "c-family/$3.c", NULL_FRULACT},
+
+  { "^(([^/]*/)*)c-family/([[:alnum:]_-]*)\\.h$",
+    REG_EXTENDED, NULL_REGEX,
+    "gt-c-family-$3.h", "c-family/$3.h", NULL_FRULACT},
+
+  /* Both c-lang.h & c-tree.h gives gt-c-decl.h for c-decl.c !  */
+  { "^(([^/]*/)*)c-lang\\.h$",
+    REG_EXTENDED, NULL_REGEX, "gt-c-decl.h", "c-decl.c", NULL_FRULACT},
+
+  { "^(([^/]*/)*)c-tree\\.h$",
+    REG_EXTENDED, NULL_REGEX, "gt-c-decl.h", "c-decl.c", NULL_FRULACT},
+
+  /* cp/cp-tree.h gives gt-cp-tree.h for cp/tree.c !  */
+  { "^(([^/]*/)*)cp/cp-tree\\.h$",
+    REG_EXTENDED, NULL_REGEX,
+    "gt-cp-tree.h", "cp/tree.c", NULL_FRULACT },
+
+  /* cp/decl.h & cp/decl.c gives gt-cp-decl.h for cp/decl.c !  */
+  { "^(([^/]*/)*)cp/decl\\.[ch]$",
+    REG_EXTENDED, NULL_REGEX,
+    "gt-cp-decl.h", "cp/decl.c", NULL_FRULACT },
+
+  /* cp/name-lookup.h gives gt-cp-name-lookup.h for cp/name-lookup.c !  */
+  { "^(([^/]*/)*)cp/name-lookup\\.h$",
+    REG_EXTENDED, NULL_REGEX,
+    "gt-cp-name-lookup.h", "cp/name-lookup.c", NULL_FRULACT },
+
+  /* objc/objc-act.h fives gt-objc-objc-act.h for objc/objc-act.c !  */
+  { "^(([^/]*/)*)objc/objc-act\\.h$",
+    REG_EXTENDED, NULL_REGEX,
+    "gt-objc-objc-act.h", "objc/objc-act.c", NULL_FRULACT },
+
+  /* General cases.  For header *.h and source *.c files, we need
+   * special actions to handle the language.  */
+
+  /* Source *.c files are using get_file_gtfilename to compute their
+     output_name and get_file_basename to compute their for_name
+     thru the source_dot_c_frul action.  */
+  { "^(([^/]*/)*)([[:alnum:]_-]*)\\.c$",
+    REG_EXTENDED, NULL_REGEX, "gt-$3.h", "$3.c", source_dot_c_frul},
+  /* Common header files get "gtype-desc.c" as their output_name,
+   * while language specific header files are handled specially.  So
+   * we need the header_dot_h_frul action.  */
+  { "^(([^/]*/)*)([[:alnum:]_-]*)\\.h$",
+    REG_EXTENDED, NULL_REGEX, "gt-$3.h", "$3.h", header_dot_h_frul},
+
+  { "^(([^/]*/)*)([[:alnum:]_-]*)\\.in$",
+    REG_EXTENDED, NULL_REGEX, "gt-$3.h", "$3.in", NULL_FRULACT},
+
+  /* Mandatory null last entry signaling end of rules.  */
+  {NULL, 0, NULL_REGEX, NULL, NULL, NULL_FRULACT}
+};
+
+/* Special file rules action for handling *.h header files.  It gives
+   "gtype-desc.c" for common headers and compute language specific
+   header files.  */
+static outf_p
+header_dot_h_frul (input_file* inpf, char**poutname, char**pforname)
+{
+  const char *basename = 0;
+  int lang_index = 0;
+  const char* inpname = get_input_file_name (inpf);
+  DBGPRINTF ("inpf %p inpname %s outname %s forname %s",
+	     (void*) inpf, inpname, *poutname, *pforname);
+  basename = get_file_basename (inpf);
+  lang_index = get_prefix_langdir_index (basename);
+  DBGPRINTF ("basename %s lang_index %d", basename, lang_index);
+
+  if (lang_index >= 0)
+    {
+      /* The header is language specific.  Given output_name &
+	 for_name remains unchanged.  The base_files array gives the
+	 outf_p.  */
+      DBGPRINTF ("header_dot_h found language specific @ %p '%s'",
+		 (void*) base_files[lang_index],
+		 (base_files[lang_index])->name);
+      return base_files[lang_index];
+    }
+  else
+    {
+      /* The header is common to all front-end languages.  So
+	 output_name is "gtype-desc.c" file.  The calling function
+	 get_output_file_with_visibility will find its outf_p.  */
+      free (*poutname);
+      *poutname = CONST_CAST (char*, "gtype-desc.c");
+      DBGPRINTF ("special 'gtype-desc.c' for inpname %s", inpname);
+      return NULL;
+    }
+}
+
+
+/* Special file rules action for handling *.c source files using
+ * get_file_gtfilename to compute their output_name and
+ * get_file_basename to compute their for_name.  The output_name is
+ * gt-<LANG>-<BASE>.h for language specific source files, and
+ * gt-<BASE>.h for common source files.  */
+static outf_p
+source_dot_c_frul (input_file* inpf, char**poutname, char**pforname)
+{
+  char *newbasename = NULL;
+  char* newoutname = NULL;
+  const char* inpname = get_input_file_name (inpf);
+  DBGPRINTF ("inpf %p inpname %s oriiginal outname %s forname %s",
+	     (void*) inpf, inpname, *poutname, *pforname);
+  newoutname = CONST_CAST (char*, get_file_gtfilename (inpf));
+  DBGPRINTF ("newoutname %s", newoutname);
+  newbasename = CONST_CAST (char*, get_file_basename (inpf));
+  DBGPRINTF ("newbasename %s", newbasename);
+  free (*poutname);
+  free (*pforname);
+  *poutname = newoutname;
+  *pforname = newbasename;
+  return NULL;
+}
+
+/* Utility function for get_output_file_with_visibility which returns
+ * a malloc-ed substituted string using TRS on matching of the FILNAM
+ * file name, using the PMATCH array.  */
+static char*
+matching_file_name_substitute (const char *filnam, regmatch_t pmatch[10],
+			       const char* trs)
+{
+  struct obstack str_obstack;
+  char* str = NULL;
+  char* rawstr = NULL;
+  const char* pt = NULL;
+  DBGPRINTF ("filnam %s", filnam);
+  obstack_init (&str_obstack);
+  for (pt = trs; *pt; pt++) {
+    char c = *pt;
+    if (c == '$')
+      {
+	if (pt[1] == '$')
+	  {
+	    /* A double dollar $$ is substituted by a single verbatim
+	       dollar, but who really uses dollar signs in file
+	       paths? */
+	    obstack_1grow (&str_obstack, '$');
+	  }
+	else if (ISDIGIT (pt[1]))
+	  {
+	    /* Handle $0 $1 ... $9 by appropriate substitution.  */
+	    int dolnum = pt[1] - '0';
+	    int so = pmatch[dolnum].rm_so;
+	    int eo = pmatch[dolnum].rm_eo;
+	    DBGPRINTF ("so=%d eo=%d dolnum=%d", so, eo, dolnum);
+	    if (so>=0 && eo>=so)
+	      obstack_grow (&str_obstack, filnam + so, eo - so);
+	  }
+	else
+	  /* This can happen only when files_rules is buggy! */
+	  fatal ("invalid dollar in transform string %s", trs);
+	/* Always skip the character after the dollar.  */
+	pt++;
+      }
+    else
+      obstack_1grow (&str_obstack, c);
+  }
+  obstack_1grow (&str_obstack, (char) 0);
+  rawstr = XOBFINISH (&str_obstack, char *);
+  str = xstrdup (rawstr);
+  obstack_free (&str_obstack, rawstr);
+  DBGPRINTF ("matched replacement %s", str);
+  rawstr = NULL;
+  return str;
+}
+
+
 
+/* An output file, suitable for definitions, that can see declarations
+   made in INPF and is linked into every language that uses INPF.  Use
+   the files_rules and regexp matching.  */
 outf_p
 get_output_file_with_visibility (input_file* inpf)
 {
-  outf_p r;
-  size_t len;
-  const char *basename;
-  const char *for_name;
-  const char *output_name;
+  outf_p r = NULL;
+  char *for_name = NULL;
+  char *output_name = NULL;
+  const char* inpfname = NULL;
 
   /* This can happen when we need a file with visibility on a
      structure that we've never seen.  We have to just hope that it's
@@ -1747,7 +1990,8 @@  get_output_file_with_visibility (input_f
     inpf = system_h_file;
 
   /* In plugin mode, return NULL unless the INPF is one of the
-     plugin_files.  We can compare input_file-s by pointer equality.  */
+     plugin_files.  We can compare input_file-s by pointer equality.
+     We don't use our files_rules machinery in plugin mode.  */
   if (plugin_files)
     {
       size_t i;
@@ -1758,69 +2002,133 @@  get_output_file_with_visibility (input_f
       return NULL;
     }
 
-  /* Determine the output file name.  */
-  basename = get_file_basename (inpf);
+  inpfname = get_input_file_name (inpf);
+  /* Use our files_rules machinery, by trying each rule in sequence
+     until a rule is triggered.  Rule regexpr-s are compiled only once,
+     lazily.  */
+  {
+    int rulix = 0;
+    DBGPRINTF ("passing input file @ %p named %s thru the files_rules",
+	       (void*) inpf, inpfname);
 
-  len = strlen (basename);
-  if ((len > 2 && memcmp (basename+len-2, ".c", 2) == 0)
-      || (len > 2 && memcmp (basename+len-2, ".y", 2) == 0)
-      || (len > 3 && memcmp (basename+len-3, ".in", 3) == 0))
-    {
-      output_name = get_file_gtfilename (inpf);
-      for_name = basename;
-    }
-  /* Some headers get used by more than one front-end; hence, it
-     would be inappropriate to spew them out to a single gtype-<lang>.h
-     (and gengtype doesn't know how to direct spewage into multiple
-     gtype-<lang>.h headers at this time).  Instead, we pair up these
-     headers with source files (and their special purpose gt-*.h headers).  */
-  else if (strncmp (basename, "c-family", 8) == 0
-	   && IS_DIR_SEPARATOR (basename[8])
-	   && strcmp (basename + 9, "c-common.h") == 0)
-    output_name = "gt-c-family-c-common.h", for_name = "c-family/c-common.c";
-  else if (strcmp (basename, "c-lang.h") == 0)
-    output_name = "gt-c-decl.h", for_name = "c-decl.c";
-  else if (strcmp (basename, "c-tree.h") == 0)
-    output_name = "gt-c-decl.h", for_name = "c-decl.c";
-  else if (strncmp (basename, "cp", 2) == 0 && IS_DIR_SEPARATOR (basename[2])
-	   && strcmp (basename + 3, "cp-tree.h") == 0)
-    output_name = "gt-cp-tree.h", for_name = "cp/tree.c";
-  else if (strncmp (basename, "cp", 2) == 0 && IS_DIR_SEPARATOR (basename[2])
-	   && strcmp (basename + 3, "decl.h") == 0)
-    output_name = "gt-cp-decl.h", for_name = "cp/decl.c";
-  else if (strncmp (basename, "cp", 2) == 0 && IS_DIR_SEPARATOR (basename[2])
-	   && strcmp (basename + 3, "name-lookup.h") == 0)
-    output_name = "gt-cp-name-lookup.h", for_name = "cp/name-lookup.c";
-  else if (strncmp (basename, "objc", 4) == 0 && IS_DIR_SEPARATOR (basename[4])
-	   && strcmp (basename + 5, "objc-act.h") == 0)
-    output_name = "gt-objc-objc-act.h", for_name = "objc/objc-act.c";
-  else
-    {
-      int lang_index = get_prefix_langdir_index (basename);
+    for (; files_rules[rulix].frul_srcexpr != NULL; rulix++)
+      {
+	DBGPRINTF ("rulix#%d srcexpr %s",
+		   rulix, files_rules[rulix].frul_srcexpr);
 
-      if (lang_index >= 0)
-	return base_files[lang_index];
+	if (!files_rules[rulix].frul_re)
+	  {
+	    /* The regexpr has not been compiled yet.  We lazily
+	       compile it only once.  */
+	    int err = 0;
+	    files_rules[rulix].frul_re = XCNEW (regex_t);
+	    err = regcomp (files_rules[rulix].frul_re,
+			   files_rules[rulix].frul_srcexpr,
+			   files_rules[rulix].frul_rflags);
+	    if (err)
+	      {
+		/* The regular expression compilation fails only when
+		   file_rules is buggy.  We give a possibly truncated
+		   error message in this impossible case, when the
+		   files_rules sequence is erroneous.  */
+		char rxerrbuf[128];
+		memset (rxerrbuf, 0, sizeof (rxerrbuf));
+		regerror (err, files_rules[rulix].frul_re,
+			  rxerrbuf, sizeof (rxerrbuf)-1);
+		fatal ("file rule #%d regexpr error %s", rulix, rxerrbuf);
+	      }
+	  }
 
-      output_name = "gtype-desc.c";
-      for_name = NULL;
-    }
+	output_name = NULL;
+	for_name = NULL;
 
-  /* Look through to see if we've ever seen this output filename before.  */
+	/* Match the regexpr and trigger the rule if matched.  */
+	{
+	  /* We have exactly ten pmatch-s, one for each $0, $1, $2,
+	     $3, ... $9.  */
+	  regmatch_t pmatch[10];
+	  memset (pmatch, 0, sizeof (pmatch));
+	  if (!regexec (files_rules[rulix].frul_re,
+			inpfname, 10, pmatch, 0))
+	    {
+	      DBGPRINTF ("input @ %p filename %s matched rulix#%d pattern %s",
+			 (void*) inpf, inpfname, rulix,
+			 files_rules[rulix].frul_srcexpr);
+	      for_name =
+		matching_file_name_substitute (inpfname, pmatch,
+					       files_rules[rulix].frul_tr_for);
+	      DBGPRINTF ("for_name %s", for_name);
+	      output_name =
+		matching_file_name_substitute (inpfname, pmatch,
+					       files_rules[rulix].frul_tr_out);
+	      DBGPRINTF ("output_name %s", output_name);
+	      if (files_rules[rulix].frul_action)
+		{
+		  /* Invoke our action routine.  */
+		  outf_p of = NULL;
+		  DBGPRINTF ("before action rulix#%d output_name %s for_name %s",
+			     rulix, output_name, for_name);
+		  of =
+		    (files_rules[rulix].frul_action) (inpf,
+						      &output_name, &for_name);
+		  DBGPRINTF ("after action rulix#%d of=%p output_name %s for_name %s",
+			     rulix, (void*)of, output_name, for_name);
+		  /* If the action routine returned something, give it back
+		     immediately and cache it in inpf.  */
+		  if (of)
+		    {
+		      inpf->inpoutf = of;
+		      return of;
+		    }
+		}
+	      /* The rule matched, and had no action, or that action did
+		 not return any output file but could have changed the
+		 output_name or for_name.  We break out of the loop on the
+		 files_rules.  */
+	      break;
+	    }
+	  else
+	    {
+	      /* The regexpr did not match.  */
+	      DBGPRINTF ("rulix#%d did not match %s pattern %s",
+			 rulix, inpfname, files_rules[rulix].frul_srcexpr);
+	      continue;
+	    }
+	}
+      }
+  }
+
+  if (!output_name || !for_name)
+    /* This is impossible, and could only happen if the files_rules is
+       incomplete or buggy.  */
+    fatal ("no output name for input file %s", inpfname);
+
+  /* Look through to see if we've ever seen this output filename
+     before.  If found, cache the result in inpf.  */
   for (r = output_files; r; r = r->next)
     if (strcmp (r->name, output_name) == 0)
-      return r;
+      {
+	inpf->inpoutf = r;
+	DBGPRINTF ("found r @ %p for output_name %s for_name %s", (void*)r,
+		   output_name, for_name);
+	return r;
+      }
 
-  /* If not, create it.  */
+  /* If not found, create it, and cache it in inpf.  */
   r = create_file (for_name, output_name);
 
   gcc_assert (r && r->name);
+  DBGPRINTF ("created r @ %p for output_name %s for_name %s", (void*) r,
+	     output_name, for_name);
+  inpf->inpoutf = r;
   return r;
 }
 
+
+
 /* The name of an output file, suitable for definitions, that can see
    declarations made in INPF and is linked into every language
    that uses INPF.  */
-
 const char *
 get_output_file_name (input_file *inpf)
 {