Integration of EXEC preprocessor beyond baseline. by utam-1 · Pull Request #294 · OCamlPro/gnucobol

utam-1 · 2026-06-14T02:38:13Z

Description of PR

This is a follow - up to #279.

Approach and basic idea

cb_preparser_entry is added to cobc.c, which handles the associated pre-parser metadata. cb_preparser_list and cb_active_preparser are created, one being a list to registered preparsers, and other is a pointer for pplex.l to identify the tag.

cb_load_preparser_conf() and cb_find_preparser() are created based on similar mechanism of cb_config_entry().
In ppparse.y a token SUBSYSTEM_FOUND is declared which is used for identifying EOF, and for gracefully exiting the parsing.

So far only general tests have been performed by using a test.conf file and setting up a pseudo - preprocessor via bash, which shows the replacement works and flag-injection is performed via gcc.

GitMensch

new changes need a new Changelog entry
we'll need a NEWS entry for the new --preparser, as well as help.c and gnucobol.texi addition; the later should also document a bit about the file format
please add your tests (you've seem to have done some) to used_binaries.at; we'd need at least:
- --preparser for not existing file
- --preparser for existing, but invalid file
- --preparser for existing file
- --preparser for two existing files
- --preparser, where pseudo preparser runs fine
- --preparser, where psuedo preparser runs in error

GitMensch · 2026-07-20T08:27:43Z

+/* Resolve preparser config filename: explicit path (contains separator
+ * or exists as-is) is used directly; otherwise look up
+ * COB_CONFIG_DIR/<name>.conf, mirroring cb_load_conf_file(). */
+static FILE *
+open_preparser_conf (const char *name, char *resolved, size_t resolved_size)
+{
+        FILE    *fp;
+        size_t  i;
+
+        for (i = 0; name[i] != 0 && name[i] != SLASH_CHAR; i++);
+
+        if (name[i] != 0 || access (name, F_OK) == 0) {
+                /* contains a path separator, or exists as given */
+                snprintf (resolved, resolved_size, "%s", name);
+        } else {
+                /* plain name: look in COB_CONFIG_DIR/<name>.conf */
+                snprintf (resolved, resolved_size, "%s%c%s.conf",
+                          cob_config_dir, SLASH_CHAR, name);
+        }
+
+        fp = fopen (resolved, "r");
+        return fp;
+}
+
+/* Parse a single "key: value" or "key value" line.
+ * Leading/trailing whitespace and comments (#) are stripped. */
+static int
+parse_preparser_line (char *buff, char **key, char **val)
+{
+        char    *p, *colon;
+
+        /* strip comment */
+        if ((p = strchr (buff, '#')) != NULL) {
+                *p = '\0';
+        }
+        /* strip trailing whitespace/newline */
+        for (p = buff + strlen (buff); p > buff && isspace ((unsigned char)p[-1]); p--);
+        *p = '\0';
+        /* strip leading whitespace */
+        for (p = buff; isspace ((unsigned char)*p); p++);
+        if (*p == '\0') {
+                return 1;       /* blank line */
+        }
+
+        *key = p;
+        if ((colon = strchr (p, ':')) != NULL) {
+                *colon = '\0';
+                p = colon + 1;
+        } else {
+                /* space-separated: key value */
+                for (; *p && !isspace ((unsigned char)*p); p++);
+                if (*p) {
+                        *p++ = '\0';
+                }
+        }
+        for (; isspace ((unsigned char)*p); p++);
+        *val = p;
+
+        /* trim trailing whitespace of key */
+        for (p = *key + strlen (*key); p > *key && isspace ((unsigned char)p[-1]); p--);
+        *p = '\0';
+
+        return 0;
+}
+
+/* Load an external preparser configuration file and register it
+ * in cb_preparser_list. Returns 0 on success, non-zero on error. */
+int
+cb_load_preparser_conf (const char *name)
+{
+        FILE                            *fp;
+        char                            resolved[COB_NORMAL_BUFF];
+        char                            buff[COB_SMALL_BUFF];
+        struct cb_preparser_entry       *pe;
+
+        fp = open_preparser_conf (name, resolved, sizeof (resolved));
+        if (!fp) {
+                cb_error (_("preparser configuration '%s' not found"), name);
+                return 1;
+        }
+
+        pe = cobc_main_malloc (sizeof (struct cb_preparser_entry));
+        pe->tag      = NULL;
+        pe->command  = NULL;
+        pe->cflags   = NULL;
+        pe->ldflags  = NULL;
+        pe->on_error = 1;       /* default: error */
+        pe->used     = 0;
+        pe->next     = NULL;
+
+        while (fgets (buff, sizeof (buff), fp)) {
+                char    *key, *val;
+
+                if (parse_preparser_line (buff, &key, &val) != 0) {
+                        continue;
+                }
+
+                if (strcasecmp (key, "tag") == 0) {
+                        size_t  i;
+                        pe->tag = cobc_main_strdup (val);
+                        for (i = 0; pe->tag[i]; i++) {
+                                pe->tag[i] = (char)toupper ((unsigned char)pe->tag[i]);
+                        }
+                } else if (strcasecmp (key, "command") == 0) {
+                        pe->command = cobc_main_strdup (val);
+                } else if (strcasecmp (key, "cflags") == 0) {
+                        pe->cflags = cobc_main_strdup (val);
+                } else if (strcasecmp (key, "ldflags") == 0) {
+                        pe->ldflags = cobc_main_strdup (val);
+                } else if (strcasecmp (key, "on-error") == 0) {
+                        pe->on_error = (strcasecmp (val, "warn") == 0) ? 0 : 1;
+                } else {
+                        cb_warning (cb_warn_unsupported,
+                                     _("unknown preparser configuration key '%s' in '%s'"),
+                                     key, resolved);
+                }
+        }
+        fclose (fp);
+
+        if (!pe->tag || !pe->command) {
+                cb_error (_("preparser configuration '%s' is missing 'tag' or 'command'"),
+                          resolved);
+                return 1;
+        }
+
+        pe->next = cb_preparser_list;
+        cb_preparser_list = pe;
+        return 0;
+}


those functions should be moved to config.c - and, where possible, the existing functions be refactored to not have duplicated code for how the lines are read/split

I kept the implementation for the two differently as they follow different kinds of implementation ( config table and Linked list ) the coupling between these two may make it difficult to maintain in my opinion.

parse_preparser_line() should be replaced by re-using the code for the config lines; to do so move the code of cb_config_entry for the parsing out of the function (the new static function gets a pointer to name+value, just as your version) - in this new variant name and value will be parsed first before cb_config_entry() checks the name (which is now intermixed).

This way we have a single new function called for both compiler and preparser config, ensuring the format is exactly identical, by keeping the way it is looked up / used out of that code.

I agree that cb_load_preparser_conf() itself should be kept separate from the config file loading (though they will look very similar) because of the struct / array approach (which we don't need fo the preparser).

GitMensch

open from last review:

new changes need a new Changelog entry (you've added them, just in the wrong place)
gnucobol.texi should also document a bit about the file format

GitMensch · 2026-07-31T16:42:58Z

 User-defined dialect configuration.

+@item --preparser=<file>
+Register external preparser configuration.


That's a start, but similar to the note below you need to explain what this is used for and either document the structure here or add an example definition (for esqlOC and/or GixSQL) showing the format with explanations.

GitMensch · 2026-07-31T16:59:35Z

+/* Resolve preparser config filename: explicit path (contains separator
+ * or exists as-is) is used directly; otherwise look up
+ * COB_CONFIG_DIR/<name>.conf, mirroring cb_load_conf_file(). */
+static FILE *
+open_preparser_conf (const char *name, char *resolved, size_t resolved_size)
+{
+        FILE    *fp;
+        size_t  i;
+
+        for (i = 0; name[i] != 0 && name[i] != SLASH_CHAR; i++);
+
+        if (name[i] != 0 || access (name, F_OK) == 0) {
+                /* contains a path separator, or exists as given */
+                snprintf (resolved, resolved_size, "%s", name);
+        } else {
+                /* plain name: look in COB_CONFIG_DIR/<name>.conf */
+                snprintf (resolved, resolved_size, "%s%c%s.conf",
+                          cob_config_dir, SLASH_CHAR, name);
+        }
+
+        fp = fopen (resolved, "r");
+        return fp;
+}
+
+/* Parse a single "key: value" or "key value" line.
+ * Leading/trailing whitespace and comments (#) are stripped. */
+static int
+parse_preparser_line (char *buff, char **key, char **val)
+{
+        char    *p, *colon;
+
+        /* strip comment */
+        if ((p = strchr (buff, '#')) != NULL) {
+                *p = '\0';
+        }
+        /* strip trailing whitespace/newline */
+        for (p = buff + strlen (buff); p > buff && isspace ((unsigned char)p[-1]); p--);
+        *p = '\0';
+        /* strip leading whitespace */
+        for (p = buff; isspace ((unsigned char)*p); p++);
+        if (*p == '\0') {
+                return 1;       /* blank line */
+        }
+
+        *key = p;
+        if ((colon = strchr (p, ':')) != NULL) {
+                *colon = '\0';
+                p = colon + 1;
+        } else {
+                /* space-separated: key value */
+                for (; *p && !isspace ((unsigned char)*p); p++);
+                if (*p) {
+                        *p++ = '\0';
+                }
+        }
+        for (; isspace ((unsigned char)*p); p++);
+        *val = p;
+
+        /* trim trailing whitespace of key */
+        for (p = *key + strlen (*key); p > *key && isspace ((unsigned char)p[-1]); p--);
+        *p = '\0';
+
+        return 0;
+}
+
+/* Load an external preparser configuration file and register it
+ * in cb_preparser_list. Returns 0 on success, non-zero on error. */
+int
+cb_load_preparser_conf (const char *name)
+{
+        FILE                            *fp;
+        char                            resolved[COB_NORMAL_BUFF];
+        char                            buff[COB_SMALL_BUFF];
+        struct cb_preparser_entry       *pe;
+
+        fp = open_preparser_conf (name, resolved, sizeof (resolved));
+        if (!fp) {
+                cb_error (_("preparser configuration '%s' not found"), name);
+                return 1;
+        }
+
+        pe = cobc_main_malloc (sizeof (struct cb_preparser_entry));
+        pe->tag      = NULL;
+        pe->command  = NULL;
+        pe->cflags   = NULL;
+        pe->ldflags  = NULL;
+        pe->on_error = 1;       /* default: error */
+        pe->used     = 0;
+        pe->next     = NULL;
+
+        while (fgets (buff, sizeof (buff), fp)) {
+                char    *key, *val;
+
+                if (parse_preparser_line (buff, &key, &val) != 0) {
+                        continue;
+                }
+
+                if (strcasecmp (key, "tag") == 0) {
+                        size_t  i;
+                        pe->tag = cobc_main_strdup (val);
+                        for (i = 0; pe->tag[i]; i++) {
+                                pe->tag[i] = (char)toupper ((unsigned char)pe->tag[i]);
+                        }
+                } else if (strcasecmp (key, "command") == 0) {
+                        pe->command = cobc_main_strdup (val);
+                } else if (strcasecmp (key, "cflags") == 0) {
+                        pe->cflags = cobc_main_strdup (val);
+                } else if (strcasecmp (key, "ldflags") == 0) {
+                        pe->ldflags = cobc_main_strdup (val);
+                } else if (strcasecmp (key, "on-error") == 0) {
+                        pe->on_error = (strcasecmp (val, "warn") == 0) ? 0 : 1;
+                } else {
+                        cb_warning (cb_warn_unsupported,
+                                     _("unknown preparser configuration key '%s' in '%s'"),
+                                     key, resolved);
+                }
+        }
+        fclose (fp);
+
+        if (!pe->tag || !pe->command) {
+                cb_error (_("preparser configuration '%s' is missing 'tag' or 'command'"),
+                          resolved);
+                return 1;
+        }
+
+        pe->next = cb_preparser_list;
+        cb_preparser_list = pe;
+        return 0;
+}


parse_preparser_line() should be replaced by re-using the code for the config lines; to do so move the code of cb_config_entry for the parsing out of the function (the new static function gets a pointer to name+value, just as your version) - in this new variant name and value will be parsed first before cb_config_entry() checks the name (which is now intermixed).

This way we have a single new function called for both compiler and preparser config, ensuring the format is exactly identical, by keeping the way it is looked up / used out of that code.

I agree that cb_load_preparser_conf() itself should be kept separate from the config file loading (though they will look very similar) because of the struct / array approach (which we don't need fo the preparser).

GitMensch · 2026-07-31T17:01:00Z

you want one entry for the baseline (with the old date) and another one for the follow-up work, which then outlines the changes in config.c, tree.h, ...

GitMensch · 2026-07-31T17:23:47Z

that whole block should - for the baseline changes - be after the may changes, nomt up-front

GitMensch · 2026-07-31T17:28:35Z

 #include <string.h>
 #include <limits.h>
-
+#include <ctype.h>


the existing code has no isspace, so you can get rid of that header when refactoring the existing code to be used for the tag/value split

GitMensch · 2026-07-31T17:29:43Z

+2026-07-31  Uttam Bhadauriya <uttamsinghbhadoriya23@gmail.com>
+
+	* cobc/cobc.h, cobc/tree.h, cobc/cobc.c, cobc/config.c, cobc/pplex.l:
+	Refactor external preparser subsystem, move struct and functions,
+	rename tag to subsystem, improve error messages and indentation.
+	* cobc/help.c, doc/gnucobol.texi: Add --preparser documentation.
+	* tests/testsuite.src/used_binaries.at: Add preparser tests.
+


cobc and doc go to their own changelog files, the testsuite sources don't get an entry for new elements

GitMensch · 2026-07-31T17:36:03Z


 * New GnuCOBOL features

+** cobc provides a --preparser option to register external preparser configurations


explain here that EXEC is now supported using external preparsers, registered with --preparser (you can add a hint to check the details in the manual, similar to the existing entries doing so) with a fallback of ignoring any EXEC parts but INCLUDE

GitMensch · 2026-07-31T17:36:50Z

 * New GnuCOBOL features

+** cobc provides a --preparser option to register external preparser configurations
+


new entries to the NEWS file are added at the end of the relevant chapter

that's still a TODO

GitMensch · 2026-07-31T17:40:54Z

+AT_CHECK([$COBC --preparser missing.conf prog.cob], [1], [], [stderr])
+AT_CHECK([$GREP "cannot load preparser configuration 'missing.conf'" stderr], [0], [ignore], [ignore])


haven't checked the ones below, but instead of placing that to an stderr where you grep from (which either needs a redirection or I learn something new here) you'd just check the output in that [] directly.

GitMensch · 2026-07-31T17:42:53Z

 AT_CLEANUP
+
+AT_SETUP([external preparser --preparser])
+AT_KEYWORDS([preparser])


Suggested change

AT_KEYWORDS([preparser])

AT_KEYWORDS([cobc configuration EXEC])

you can replace the keyword by cobc + configuration - other words are already included by AT_SETUP content

GitMensch · 2026-08-05T12:14:45Z

@utam-1 please rebase, which should improve the MSVC part and update per review, so we can have a next look at that

MacOS: pass std=gnu17 MSYS2: disable screenio tests for now overall actions update (fix npm deprecation)

cobc/: * config.def: "tab-width" option is changed to a comma-separated list of tab widths, for example "6,1,4", the last one being reused. This new meaning is backward compatible. Implements FR #498 " Adjust tab-width to optionally be a list of tab-stop positions" * config.c: initialize cb_tab_width as a string, each position indicating the number of spaces to insert for a tab at that position * pplex.c,cobc.c: use the new type of cb_tab_width

cobc: * tree.h, parser.y: change type of cobc_cs_check flags to permit distinguishing more grammar contexts * cobc.h, parser.y, reserved.c (lookup_reserved_word): generalize and cleanup device for handling special contexts * parser.y: use new macros __CS_ENSURE, __CS_CLEAR, __CS_CLEAR_ALL, __CS_CHECK, __CS_LEAVE, __CS_ENTER, to help debug and explicitly manipulate special contexts * error.c (cb_error_always, cb_error_internal), tree.h, tree.c (cb_build_program), scanner.l: replace cobc_in_repository flag with a check on a special context * scanner.l: introduce cobc_in_exit_statement helper macro

utam-1 · 2026-08-06T04:51:39Z

@utam-1 please rebase, which should improve the MSVC part and update per review, so we can have a next look at that

I have pushed my changes, and apologies for the delay I was a bit occupied.

cobc/: * config.def: "tab-width" option is changed to a comma-separated list of tab widths, for example "6,1,4", the last one being reused. This new meaning is backward compatible. Implements FR #498 " Adjust tab-width to optionally be a list of tab-stop positions" * config.c: initialize cb_tab_width as a string, each position indicating the number of spaces to insert for a tab at that position * pplex.c,cobc.c: use the new type of cb_tab_width

cobc: * tree.h, parser.y: change type of cobc_cs_check flags to permit distinguishing more grammar contexts * cobc.h, parser.y, reserved.c (lookup_reserved_word): generalize and cleanup device for handling special contexts * parser.y: use new macros __CS_ENSURE, __CS_CLEAR, __CS_CLEAR_ALL, __CS_CHECK, __CS_LEAVE, __CS_ENTER, to help debug and explicitly manipulate special contexts * error.c (cb_error_always, cb_error_internal), tree.h, tree.c (cb_build_program), scanner.l: replace cobc_in_repository flag with a check on a special context * scanner.l: introduce cobc_in_exit_statement helper macro

Credits goes to Saurabh Kumar for the use of vswhere to locate the Visual Studio installation. Co-authored-by: Saurabh Kumar <developer.saurabh@outlook.com>

utam-1 · 2026-08-06T14:49:32Z

@GitMensch I think the issue with msvc persists, maybe I made some mistake while doing the rebase. I'm not sure though.

GitMensch · 2026-08-06T14:46:02Z

+# 1. nonexistent file -> error
+AT_CHECK([$COBC --preparser ./missing.conf prog.cob], [1], [],
+[configuration error:
+./missing.conf: No such file or directory
+])
+
+# 2. invalid file -> error
+AT_DATA([invalid.conf], [
+# missing subsystem and command
+cflags: -O2
+])
+AT_CHECK([$COBC --preparser ./invalid.conf prog.cob], [1], [],
+[error: preparser configuration './invalid.conf' is missing 'subsystem' or 'command'
+])


I suggest to move these parts to configuration.at and add the missing "unknown configuration tag" code path

GitMensch · 2026-08-06T15:07:51Z

+		/* Simplified: Always append .conf for plain names in config dir */
+		snprintf (resolved, resolved_size, "%s%c%s.conf",
+			  cob_config_dir, SLASH_CHAR, name);


this simplification is likely the reason for the win32 failures:

you want to only do the fopen below if the access worked

the snprintf should only be done if the access worked as you want to point to the "normal" name if there is neither one with the original name nor with the config dir

I made this change but it seems to fail again for some reason.

you still set resolved_size if the path did not exist and use it afterwards, I'll do an edit you can inspect and we'll see if that helps

This reverts commit ed29d25.

utam-1 · 2026-08-07T05:33:20Z

@GitMensch The tests are still failing, I'm not sure what's causing this from the error messages it says about some dialect of C89 / C11 not being met ( for declartion of for-loop ) and some other path related issues are present.

GitMensch · 2026-08-07T09:48:22Z

@GitMensch The tests are still failing, I'm not sure what's causing this from the error messages it says about some dialect of C89 / C11 not being met ( for declartion of for-loop ) and some other path related issues are present.

That's

../../cobc/config.c: In function ‘cb_conf_split_line’:
../../cobc/config.c:938:17: error: ‘for’ loop initial declarations are only allowed in C99 or C11 mode
  938 |                 for (size_t j = strlen (buff);
      |                 ^~~
../../cobc/config.c:938:17: note: use option ‘-std=c99’, ‘-std=gnu99’, ‘-std=c11’ or ‘-std=gnu11’ to compile your code

and it is nice to see that the message got much better over the years :-)

It means "for" is a statement, and "size_t j" is a declaration. In C89 (the compatibility check done here) declarations need to be at the start of a block, so that rule is broken. The fix here is to move the size_t declaration before the for statement - but we don't need that in any case as we can directly adjust the pointer buff.
I'll include that in my commit.

utam-1 · 2026-08-07T13:52:18Z

@GitMensch The tests are still failing, I'm not sure what's causing this from the error messages it says about some dialect of C89 / C11 not being met ( for declartion of for-loop ) and some other path related issues are present.

That's
../../cobc/config.c: In function ‘cb_conf_split_line’:
../../cobc/config.c:938:17: error: ‘for’ loop initial declarations are only allowed in C99 or C11 mode
  938 |                 for (size_t j = strlen (buff);
      |                 ^~~
../../cobc/config.c:938:17: note: use option ‘-std=c99’, ‘-std=gnu99’, ‘-std=c11’ or ‘-std=gnu11’ to compile your code
and it is nice to see that the message got much better over the years :-)

It means "for" is a statement, and "size_t j" is a declaration. In C89 (the compatibility check done here) declarations need to be at the start of a block, so that rule is broken. The fix here is to move the size_t declaration before the for statement - but we don't need that in any case as we can directly adjust the pointer buff. I'll include that in my commit.

This is rather interesting; I'll review (and learn) from the commit for both the fix related to MSVC build and this issue once it is applied.

GitMensch · 2026-08-07T15:52:46Z

work will still take some time, so I did an untested (not even compiled so far) commit - if it fails it should still show the idea...

edit: seems you only need to update the expected test result to let it pass;
@utam-1 if you have any questions on the changes we can schedule a quick meeting

GitMensch · 2026-08-07T15:53:24Z

+	  functions to config.c; share line-parsing with cb_config_entry;
+	  fix indentation; fix cmd_len calculation
+	* help.c: add --preparser option documentation
 2026-06-08  Nicolas Berthier <nicolas.berthier@ocamlpro.com>


Suggested change

2026-06-08 Nicolas Berthier <nicolas.berthier@ocamlpro.com>

2026-06-08 Nicolas Berthier <nicolas.berthier@ocamlpro.com>

GitMensch · 2026-08-07T15:54:42Z

+	* config.c: initialize cb_tab_width as a string,
+	  each position indicating the number of spaces to insert for a tab
+	  at that position
+	* pplex.c,cobc.c: use the new type of cb_tab_width


Suggested change

* pplex.c,cobc.c: use the new type of cb_tab_width

* pplex.l, cobc.c: use the new type of cb_tab_width

... but I'm a but confused where that entry comes from...

utam-1 · 2026-08-08T15:16:46Z

work will still take some time, so I did an untested (not even compiled so far) commit - if it fails it should still show the idea...

edit: seems you only need to update the expected test result to let it pass; @utam-1 if you have any questions on the changes we can schedule a quick meeting

@GitMensch I was able to get an insight on the approach, especially for the fix related to movement of pointer in buff instead of relying at size_t. However, I believe the issue related to MSVC is still there no?
Edit: The tests are passing successfully now, I modified the tests and changed the path look-up. Let me know if any further changes are needed.

GitMensch

The overall tests and code look good - have you already run it sucessfully with the gixsql examples running gixsql?

If yes, please add the sample configuration for this (and possibly others like esqloc and esql) to the config folder.

If not: is there anything missing but the testing?

There are 3 main issues where I've asked (or suggested) for changes, I hope they are explained well enough - if not, feel free to ask here directly.

GitMensch · 2026-08-09T16:45:15Z

 * New GnuCOBOL features

+** cobc provides a --preparser option to register external preparser configurations
+


that's still a TODO

GitMensch · 2026-08-09T16:47:15Z

+prog.cob:6: error: EXEC SQL statement ignored
+])
+
+AT_CLEANUP


Suggested change

AT_CLEANUP

AT_CLEANUP

don't ask me why but some diff tools work better if there's an empty line after the cleanup...

GitMensch · 2026-08-09T16:50:56Z

+AT_CHECK([$COMPILE_ONLY -Werror=unsupported -ffold-copy=LOWER prog.cob], [1], [],
+[prog.cob:8: error: EXEC SQL INCLUDE handled as COPY
+])


Suggested change

AT_CHECK([$COMPILE_ONLY -Werror=unsupported -ffold-copy=LOWER prog.cob], [1], [],

[prog.cob:8: error: EXEC SQL INCLUDE handled as COPY

])

AT_CHECK([$COMPILE_ONLY -Wno-unsupported -ffold-copy=LOWER prog.cob], [0], [], [])

we want to verify that the copy is read, so no reason for an error

GitMensch · 2026-08-09T16:51:36Z

+AT_CHECK([$COMPILE_ONLY -Werror=unsupported -ffold-copy=LOWER prog.cob], [1], [],
+[prog.cob:8: error: EXEC SQL INCLUDE handled as COPY
+])
+


Suggested change

AT_CHECK([$COMPILE_ONLY -Werror=unsupported -ffold-copy=LOWER prog.cob], [1], [],

[prog.cob:8: error: EXEC SQL INCLUDE handled as COPY

])

AT_CHECK([$COMPILE_ONLY -Wno-unsupported -ffold-copy=LOWER prog.cob], [0], [], [])

same as above

GitMensch · 2026-08-09T16:53:03Z

+AT_CHECK([$COMPILE_ONLY -Werror=unsupported prog.cob], [1], [],
+[prog.cob:9: error: missing END-EXEC at end of file
+prog.cob:9: error: EXEC SQL statement ignored
+])


Suggested change

AT_CHECK([$COMPILE_ONLY -Werror=unsupported prog.cob], [1], [],

[prog.cob:9: error: missing END-EXEC at end of file

prog.cob:9: error: EXEC SQL statement ignored

])

AT_CHECK([$COMPILE_ONLY -Wunsupported prog.cob], [1], [],

[prog.cob:9: error: missing END-EXEC at end of file

prog.cob:9: warning: EXEC SQL statement ignored

])

no need to error that explicit - but the missing part should be an error in any case

GitMensch · 2026-08-09T18:11:57Z

+		cmd_len = strlen (cb_active_preparser->command) + strlen (fn->source) + strlen (fn->preprocess) + 3;
+		cmd = cobc_malloc (cmd_len);


move those two line directly to the declaration (making cmd_len a const and reorder it)

GitMensch · 2026-08-09T18:12:45Z

+		char *cmd;
+		size_t cmd_len;
+		int ret_sys;
+		const char *new_preprocess;


drop that var, assigning it directly to fn->preprocess

GitMensch · 2026-08-09T18:15:27Z

+			cb_source_file = fn->source;
+			if (!cb_active_preparser->warn_only) {
+				cobc_err_exit (_("external preparser '%s' failed with exit status %d"), cb_active_preparser->subsystem, ret_sys);
+			} else {


no need for the else branch, the err_exit stops the program so everything in an else can be directly placed in the code after the closing }

GitMensch · 2026-08-09T18:17:12Z

+				fn->source = orig_source;     /* Restore the original file names */
+				fn->preprocess = orig_preprocess;


Not checked, just wondering: have we overridden those vars at all? Aren't they still fine?

GitMensch · 2026-08-09T18:19:02Z

+		/*This prevents collision in file paths based on manual tests*/
+		new_preprocess = file_replace_extension ((char *)fn->preprocess, ".i2");
+
+		fn->source     = cobc_strdup (fn->preprocess);   /* sqlpp.sh's output */
+		fn->preprocess = new_preprocess;                  /* fresh file for pass 2 */
+
+		if (cb_active_preparser->cflags) {
+			COBC_ADD_STR (cobc_cflags, " ", cb_active_preparser->cflags, NULL);
+		}
+		if (cb_active_preparser->ldflags) {
+			COBC_ADD_STR (cobc_ldflags, " ", cb_active_preparser->ldflags, NULL);
+		}


this can be the else branch for the failure above... actually better check if (ret_sys )= 0) { and have that in the first branch, the hopefully less likely error case in the else - and finally the goto outside of both

GitMensch · 2026-08-09T18:35:51Z

+		e = buff + strlen (buff);
+		while (e > buff && (e[-1] == '\r' || e[-1] == '\n')) {
+			*--e = 0;
+		}


That's slightly confusing - Is there a reason to not subtract 1 when computing e and checking e directly instead of e[-1}?
You can then also just do e--; in the loop and end with a final *(e + 1) = 0; before the return.

... just wondering.

My first thoughts were to use something cleaner and readable as you suggested, but from what I gathered the subtract 1 rule results in an undefined behaviour ( when buff is empty ) as that'll make e point to an address before the start of array. Modern architectures do not take this thing into account but maybe for stricter architectures it may result in issues.
Be as it may, I have updated the code, as per your suggestions.

from what I gathered the subtract 1 rule results in an undefined behaviour ( when buff is empty )

Just interested to follow your thought: How can that be with the previous code (skipping empty lines, and the separator not found)?

You're right the previous lines do ensure that buff doesn't stay empty, but then again there are issues with aggressive compiler optimization if we're using Undefined Behaviour code, as they're picked up as warnings of some sort in old school systems, apart from that if someone includes a static analysis tools do tend to flag such codes if they're present. I personally would not vouch for encountering issues with compiler optimization when using such kind of code, however I have witnessed issues with static analysis tools complaining if we're moving away from pre-defined standards.
This is nitpicking in my humble opinion, the code as such right now wouldn't create any issues.

utam-1 · 2026-08-12T05:11:45Z

have you already run it sucessfully with the gixsql examples running gixsql?

If yes, please add the sample configuration for this (and possibly others like esqloc and esql) to the config folder.

If not: is there anything missing but the testing?

@GitMensch I have performed a crude test with GixSQL, but I believe since it wasn't rigorous enough I didn't added the config file. I am describing what I have found below:
The config file I am using looks something like this :

subsystem:  SQL
command:    /tmp/gixpp-wrapper.sh
on-error:   error

I have used the following command to invoke cobc --preparser :

cobc --preparser /tmp/sql-gixsql.conf \
     --save-temps -x \
     -I /home/user/gixsql-1.0.20b/examples \
     -I /home/user/gixsql-1.0.20b/copy \
     /home/user/gixsql-1.0.20b/examples/TSQL001A.cbl \
     -o /tmp/TSQL001A

And from this the output I got was :

GIXSQL*    EXEC SQL
GIXSQL*       CONNECT TO :DATASRC USER :DBUSR USING :DBPWD
GIXSQL*    END-EXEC.      
GIXSQL     CALL "GIXSQLConnect" USING
GIXSQL         BY REFERENCE SQLCA
GIXSQL         BY REFERENCE DATASRC
GIXSQL         BY VALUE 64
GIXSQL         BY REFERENCE x"00"
GIXSQL         BY VALUE 0
GIXSQL         BY REFERENCE x"00"
GIXSQL         BY VALUE 0
GIXSQL         BY REFERENCE DBUSR
GIXSQL         BY VALUE 64
GIXSQL         BY REFERENCE DBPWD
GIXSQL         BY VALUE 64
GIXSQL     END-CALL.

I think the translation works, however, if possible, I think it would be better if you would perform more rigorous testing from your end.

Apart from the testing, I don't recall there is anything else, as we have implemented almost everything based on the plan we had outlined.

GitMensch

please attach your LLM findings about nested preprocessor issues to this PR

Have you tried it locally (for example with a pseudo EXEC CICS which just replaces CICS by SQL and your GixSQL test configuration)?

GitMensch · 2026-08-17T19:40:21Z

+			/* These options were all processed in the first getopt-run */
+			break;


Suggested change

/* These options were all processed in the first getopt-run */

break;

GitMensch · 2026-08-17T19:44:47Z

+		/* strip trailing CR/LF for error message */
+		e = buff + strlen (buff) - 1;
+		while (e >= buff && (*e == '\r' || *e == '\n')) {
+			e--;
+		}
+		*(e + 1) = 0;


interesting side question: what happens if the file has no trailing CR/LF and we get here?

That wouldn't be in an issue I think, the loop would exit and the null terminator would be overwritten. The string wouldn't change.

GitMensch · 2026-08-17T19:50:52Z

+		line++;
+		split_ret = cb_conf_split_line (buff, &tag_name, &tag_val);
+		if (split_ret == 1) {
+			continue;


Suggested change

continue;

continue; /* blank or comment line */

GitMensch · 2026-08-17T19:56:00Z

+			configuration_error (name, line, 1,
+				_("invalid configuration tag '%s'"), buff);
+			status = 1;
+			continue;


according to the coverage check, this one misses a test case

GitMensch · 2026-08-17T20:08:13Z

+AT_DATA([unknown.conf], [
+subsystem: SQL
+command: /bin/true
+badkey: something
+])
+
+AT_DATA([prog.cob], [
+       IDENTIFICATION DIVISION.
+       PROGRAM-ID. prog.
+       PROCEDURE DIVISION.
+           STOP RUN.
+])
+
+AT_CHECK([$COBC --preparser ./unknown.conf prog.cob], [1], [],
+[configuration error:
+./unknown.conf:4: unknown configuration tag 'badkey'
+])


Suggested change

AT_DATA([unknown.conf], [

subsystem: SQL

command: /bin/true

badkey: something

])

AT_DATA([prog.cob], [

IDENTIFICATION DIVISION.

PROGRAM-ID. prog.

PROCEDURE DIVISION.

STOP RUN.

])

AT_CHECK([$COBC --preparser ./unknown.conf prog.cob], [1], [],

[configuration error:

./unknown.conf:4: unknown configuration tag 'badkey'

])

AT_DATA([unknown.conf], [

subsystem: SQL

command: /bin/true # value does not matter, we check for the key below

# (and here for inline parsing of comments)

badkey: something

])

AT_DATA([prog.cob], [

IDENTIFICATION DIVISION.

PROGRAM-ID. prog.

PROCEDURE DIVISION.

STOP RUN.

])

AT_CHECK([$COBC --preparser ./unknown.conf prog.cob], [1], [],

[configuration error:

./unknown.conf:6: unknown configuration tag 'badkey'

])

GitMensch · 2026-08-17T20:12:57Z

+		const size_t	cmd_len = strlen (cb_active_preparser->command) + strlen (fn->source) + strlen (fn->preprocess) + 3;
+		char	*cmd = cobc_malloc (cmd_len);
+		snprintf (cmd, cmd_len, "%s %s %s", cb_active_preparser->command, fn->source, fn->preprocess);
+
+		ret_sys = call_system (cmd);
+		cobc_free (cmd);


I do wonder if/how we should pass include paths explicit given to cobc, maybe just as a third long argument that contains -I path1 -I /path/two?
That still allows to do "something else" in the preparser config if -I is not supported by the preparser.

GitMensch · 2026-08-17T20:15:45Z

 #define YYSTYPE			cb_tree
 #define _PARSER_H	/* work around bad Windows SDK header */
 #include "parser.h"
+#include <stdio.h>


Suggested change

#include <stdio.h>

We don't need that as there is no new code, do we?

GitMensch · 2026-08-17T20:16:24Z

 #include "cobc.h"
 #include "tree.h"
 #include "ppparse.h"
+#include <stdio.h>


What do we need this for?

GitMensch · 2026-08-17T20:20:30Z

+Note that @code{INCLUDE} statements (e.g., @code{EXEC SQL INCLUDE ...}) are 
+always handled by the internal preprocessor as a standard @code{COPY} statement, 
+with full support for library paths (@option{-I}) and case folding, before any 
+external preparser is considered.


I think that does only apply if no preparser is found:

If a preparser is configured correctly then we would get EXEC (EXEC_STATE) WORD + SUBSYSTEM_FOUND = YYACCEPT + restart --> then the configured EXEC preparser would already see and replace INCLUDE, so COBOL would never see that in the second process, no?

utam-1 · 2026-08-19T14:21:48Z

please attach your LLM findings about nested preprocessor issues to this PR

Have you tried it locally (for example with a pseudo EXEC CICS which just replaces CICS by SQL and your GixSQL test configuration)?

@GitMensch I have added a folder called test_preparser which details the findings. I have tested this locally. The issue was with name collision in file_replace_extension which seems to have been resolved.

…eline

utam-1 added 9 commits March 16, 2026 07:35

Added initial EXEC preprocessor handling logic in pplex and ppparse

9a895d7

Modified approach based on review comments

119e265

Deleted unreleated files, and fixed inconsistencies

e492b07

Refined the test cases based on feedback

31c0601

Applied the changes based on feedback

a946047

Modified test cases and added support for EOF

88a2960

Added new test cases and subsytem token

b645b64

Restored to previous commit with modified test cases

632cc07

Modified subsystem handling for dotted names and improved test cases

5909d70

utam-1 marked this pull request as draft June 14, 2026 02:38

Implementation of step 4, 5 and 6 after baseline integration.

f037357

utam-1 marked this pull request as ready for review June 29, 2026 14:40

GitMensch requested changes Jul 20, 2026

View reviewed changes

GitMensch reviewed Jul 31, 2026

View reviewed changes

GitMensch requested changes Jul 31, 2026

View reviewed changes

GitMensch and others added 7 commits August 6, 2026 09:36

windows-msvc: fix env var reference for dependencies

3f1ef75

workflow updates

0a8d46a

MacOS: pass std=gnu17 MSYS2: disable screenio tests for now overall actions update (fix npm deprecation)

Rename is_test to cob_is_test in libcob as it is an extern value

4694240

Performed modifications based on feedback.

6ca36c7

Refactored config parser approach and modified test expectations

4054ab6

utam-1 force-pushed the exec-preprocessor-beyond-baseline branch from 6581475 to 4054ab6 Compare August 6, 2026 04:50

lefessan and others added 3 commits August 6, 2026 19:10

Fix and update CI

7b34538

Credits goes to Saurabh Kumar for the use of vswhere to locate the Visual Studio installation. Co-authored-by: Saurabh Kumar <developer.saurabh@outlook.com>

utam-1 force-pushed the exec-preprocessor-beyond-baseline branch from c9fe3f3 to 7b34538 Compare August 6, 2026 13:58

GitMensch requested changes Aug 6, 2026

View reviewed changes

utam-1 added 3 commits August 7, 2026 07:16

Placed tests in config.at and modified approach for msvc error

ed29d25

Revert "Placed tests in config.at and modified approach for msvc error"

546813b

This reverts commit ed29d25.

Changed position of test cases and updated preparser conf

126eb25

utam-1 force-pushed the exec-preprocessor-beyond-baseline branch 2 times, most recently from caafec2 to c7aa539 Compare August 7, 2026 04:44

GitMensch requested changes Aug 7, 2026

View reviewed changes

utam-1 force-pushed the exec-preprocessor-beyond-baseline branch 2 times, most recently from eaa9ed4 to 37fdedc Compare August 8, 2026 14:28

utam-1 force-pushed the exec-preprocessor-beyond-baseline branch from 6f664b0 to c3c0a87 Compare August 9, 2026 01:44

GitMensch requested changes Aug 9, 2026

View reviewed changes

GitMensch reviewed Aug 9, 2026

View reviewed changes

utam-1 force-pushed the exec-preprocessor-beyond-baseline branch 2 times, most recently from efacb58 to 8a9ae31 Compare August 12, 2026 04:06

GitMensch requested changes Aug 17, 2026

View reviewed changes

utam-1 force-pushed the exec-preprocessor-beyond-baseline branch from 9780605 to fa073ec Compare August 19, 2026 13:49

Modified common.c to resolve the build errors.

0058c20

utam-1 force-pushed the exec-preprocessor-beyond-baseline branch from f0b8c93 to 0058c20 Compare August 20, 2026 01:54

Merge branch 'gitside-gnucobol-3.x' into exec-preprocessor-beyond-bas…

b9f9d96

…eline


		* New GnuCOBOL features

		** cobc provides a --preparser option to register external preparser configurations

		AT_CHECK([$COBC --preparser missing.conf prog.cob], [1], [], [stderr])
		AT_CHECK([$GREP "cannot load preparser configuration 'missing.conf'" stderr], [0], [ignore], [ignore])

	AT_KEYWORDS([preparser])
	AT_KEYWORDS([cobc configuration EXEC])

	2026-06-08 Nicolas Berthier <nicolas.berthier@ocamlpro.com>

	2026-06-08 Nicolas Berthier <nicolas.berthier@ocamlpro.com>

	* pplex.c,cobc.c: use the new type of cb_tab_width
	* pplex.l, cobc.c: use the new type of cb_tab_width

		cmd_len = strlen (cb_active_preparser->command) + strlen (fn->source) + strlen (fn->preprocess) + 3;
		cmd = cobc_malloc (cmd_len);

		fn->source = orig_source; /* Restore the original file names */
		fn->preprocess = orig_preprocess;

		/* These options were all processed in the first getopt-run */
		break;

Uh oh!

Conversation

utam-1 commented Jun 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description of PR

Approach and basic idea

Uh oh!

GitMensch left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

GitMensch left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

GitMensch commented Aug 5, 2026

Uh oh!

utam-1 commented Aug 6, 2026 • edited by GitMensch Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

utam-1 commented Aug 6, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

utam-1 commented Aug 7, 2026

Uh oh!

GitMensch commented Aug 7, 2026

Uh oh!

utam-1 commented Aug 7, 2026

Uh oh!

GitMensch commented Aug 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

utam-1 commented Jun 14, 2026 •

edited

Loading

GitMensch left a comment •

edited

Loading

utam-1 commented Aug 6, 2026 •

edited by GitMensch

Loading

GitMensch commented Aug 7, 2026 •

edited

Loading

utam-1 commented Aug 8, 2026 •

edited

Loading

utam-1 Aug 12, 2026 •

edited

Loading