modifying ld options in the enviroment

modifying ld options in the enviroment

[Marty Leisner]

[-- Attachment #1: Type: text/plain, Size: 1317 bytes --]


I've often tried to do things like 
	make CC='gcc -Wl,-t'
to find out why builds were doing strange things -- in complicated
situations this doesn't work for a variety of reasons (depending
on how the system was architected).

Instead of rearchitecting the make architecture, or seeing the
failing the command line (which is sometimes not shown) and running it 
standalone,
I inserted a environemnt variable 
	LD_ENV_OPTIONS
which allows you to inject options into ld
(another approach is to write a script in front of the real ld which does
this, but this gets old fast)...

A simple case is:
: leisner@dell3;LD_ENV_OPTIONS=-t gcc foo.c
option is -t
/usr/gnu/binutils/bin/ld: mode elf_i386
/usr/lib/gcc/i386-redhat-linux/3.4.2/../../../crt1.o
/usr/lib/gcc/i386-redhat-linux/3.4.2/../../../crti.o
/usr/lib/gcc/i386-redhat-linux/3.4.2/crtbegin.o
/tmp/ccOGmbte.o
-lgcc_s (/usr/lib/gcc/i386-redhat-linux/3.4.2/libgcc_s.so)
/lib/libc.so.6
(/usr/lib/libc_nonshared.a)elf-init.oS
-lgcc_s (/usr/lib/gcc/i386-redhat-linux/3.4.2/libgcc_s.so)
/usr/lib/gcc/i386-redhat-linux/3.4.2/crtend.o
/usr/lib/gcc/i386-redhat-linux/3.4.2/../../../crtn.o

This is wrt 2.16 -- I've had this for a few years...I've found it 
very useful...I'm thinking of duplicating something like this for gcc...



Marty Leisner
leisner@rochester.rr.com


[-- Attachment #2: ldmain.c.diff --]
[-- Type: text/plain , Size: 2324 bytes --]

--- ldmain.c	2005/05/12 05:49:05	1.1
+++ ldmain.c	2005/05/14 20:33:57
@@ -164,6 +164,10 @@
 };
 
 struct bfd_link_info link_info;
+
+static int modify_env_argc(int argc, char ***);
+
+ 
 \f
 static void
 remove_output (void)
@@ -320,12 +324,14 @@
   config.magic_demand_paged = TRUE;
   config.text_read_only = TRUE;
 
+  argc = modify_env_args(argc, &argv);
   emulation = get_emulation (argc, argv);
   ldemul_choose_mode (emulation);
   default_target = ldemul_choose_target (argc, argv);
   lang_init ();
   ldemul_before_parse ();
   lang_has_input_file = FALSE;
+ 
   parse_args (argc, argv);
 
   if (config.hash_table_size != 0)
@@ -1526,3 +1532,93 @@
 
   return TRUE;
 }
+
+static int count_options(const char *env)
+{
+	int i;
+
+	i = 0;
+
+	while(*env) {
+		/* skip white */
+		while(*env == ' ')
+			env++;
+
+		if(*env) 
+			i++;
+
+		/* find white */
+		while(*env && *env != ' ' )
+			env++;
+	}
+	return i;
+}
+
+/* malloc space and figure out option, copy it into the space */
+static char *get_env_option(char *env)
+{
+	char *end;
+	int size;
+	char *new;
+
+	for(end = env; *end && *end != ' '; end++)
+		;
+
+	size = end - env + 1;
+	new = malloc(size);
+	strncpy(new, env, size - 1);
+	new[size - 1] = 0;
+#if 0
+	printf("option is %s\n", new);
+#endif
+	return new;
+}	
+
+static char *skip_white(char *env)
+{
+	while(*env && *env == ' ')
+		env++;
+
+	return env;
+}
+
+/* possibly modify the argc/argv list from an environment
+ * variable.
+ * If the list is modified, new arguments are inserts started
+ * at position argv[1] and everything else is shifted down.
+ * New space is allocated for the argv array.
+ */
+static int modify_env_args(int argc, 
+		           char ***pass_argv)
+{
+	char *env;
+	int count;
+	int space_needed;
+	char *cp;
+	char **new_argv;
+	char *space;
+	int i;
+	char **argv;
+
+	env = getenv("LD_ENV_OPTIONS");
+	if(!env)
+		return argc;
+
+	argv = *pass_argv;
+	count = count_options(env);
+	new_argv = malloc((argc + count + 1) * sizeof(char *));
+	new_argv[0] = argv[0];
+	for(i = 0; i < count; i++) {
+		env = skip_white(env);
+		new_argv[i + 1] = get_env_option(env);
+		env = index(env, ' ');
+	}
+	
+	/* move rest of argv to new_argv */
+	memcpy(new_argv + (i + 1), argv + 1, sizeof(char *) * argc);
+
+	*pass_argv = new_argv;
+
+	return argc + count;
+}
+	

Re: modifying ld options in the enviroment

[Nick Clifton]

Hi Marty,

> I inserted a environemnt variable 
> 	LD_ENV_OPTIONS
> which allows you to inject options into ld

> This is wrt 2.16 -- I've had this for a few years...I've found it 
> very useful...I'm thinking of duplicating something like this for gcc...

The big problem with this approach, and the reason why it has been 
rejected before (both for ld and gcc) is that it makes reproducing a bug 
much harder.  Unless you remember that the environment variable has been 
set, (assuming that you know about it, quite often this kind of thing is 
done for you in a system wide initialisation script), then you might not 
report it when submitting a bug report or take it into consideration 
when investigating a bug.

Using wrapper scripts also has the same problem.  By far the best way is 
to spend the effort to create a good build environment where extra 
linker command line switches can be set on the build command line.

Cheers
   Nick