Handling multiple -include directives
RFC PATCH 1/1 Handling multiple -include directives
Hello!
sparse doesn't handle multiple -include directives. Variables "include" and "include_fd" have file scope in lib.c. The file for inclusion is processed only once in sparse_initial() after the command line has been processed.
It seems to me that the existing add_pre_buffer() mechanism can be used instead. I'm just a bit worried why it wasn't done like this in the first place.
---
Handle multiple -include directives
---
lib.c | 26 ++++++-------------------- 1 files changed, 6 insertions(+), 20 deletions(-)
lib.c +6 -20
@@ -191,8 +191,6 @@ int Wenum_mismatch = 1;int Wdo_while =1 ;int Wuninitialized =1 ;int preprocess_only; -char *include; -int include_fd =-1 ;void add_pre_buffer (const char *fmt, ...)
@@ -266,21 +264,15 @@ static char **handle_switch_i(char *arg, char **next) {if (*next && !strcmp (arg,"include" )) {char *name = *++next; -int fd =open (name,O_RDONLY ); - - include_fd = fd; - include = name; -if (fd <0 ) -perror (name); +if (!name) +die ("missing argument for -include option" ); +add_pre_buffer ("#include \"%s\"\n" , name); }if (*next && !strcmp (arg,"imacros" )) {char *name = *++next; -int fd =open (name,O_RDONLY ); - - include_fd = fd; - include = name; -if (fd <0 ) -perror (name); +if (!name) +die ("missing argument for -include option" ); +add_pre_buffer ("#include \"%s\"\n" , name); }else if (*next && !strcmp (arg,"isystem" )) {char *path = *++next;
@@ -624,12 +616,6 @@ static struct symbol_list *sparse_initial(void) {struct token *token; -// Prepend any "include" file to the stream. -// We're in global scope, it will affect all files! - token =NULL ; -if (include_fd >=0 ) - token =tokenize (include, include_fd,NULL , includepath); -// Prepend the initial built-in stream token =tokenize_buffer (pre_buffer, pre_buffer_size, token);return sparse_tokenstream (token);
Signature · 2 lines
Regards, Pavel Roskin
Quoted 3 lines · Pavel Roskin
It seems to me that the existing
add_pre_buffer()mechanism can be used instead. I'm just a bit worried why it wasn't done like this in the first place.
There was a reason to worry. Now create_builtin_stream() is run after the includes have been processed, so that e.g. the Linux compiler.h tells me that my compiler is too old (because it was included from the command line before __GNUC__ was defined).
The initialization order needs some untangling. It's too intertwined with the command line processing.
Signature · 2 lines
Regards, Pavel Roskin
Christopher Li
maintainerQuoted 4 lines · Pavel Roskin
Quoted 3 lines · Pavel Roskin
It seems to me that the existing
add_pre_buffer()mechanism can be used instead. I'm just a bit worried why it wasn't done like this in the first place.
I believe the reason it is not in the pre buffer is that it should first search the current directory instead of the source file directory. The command line -include has some subtle differences with #include "filename"
Quoted 5 lines
There was a reason to worry. Now
create_builtin_stream()is run after the includes have been processed, so that e.g. the Linuxcompiler.htells me that my compiler is too old (because it was included from the command line before__GNUC__was defined).
That is the other reason as well :-)
Can you please try this patch and see if it works for you?
Chris
Index: sparse/pre-process.c
===================================================================
lib.c +26 -21
@@ -192,7 +192,10 @@ int Wdo_while = 1;int Wuninitialized =1 ;int preprocess_only;char *include; -int include_fd =-1 ; + +#define CMDLINE_INCLUDE 20 +int cmdline_include_nr =0 ; +struct cmdline_include cmdline_include[CMDLINE_INCLUDE ];void add_pre_buffer (const char *fmt, ...)
@@ -262,26 +265,26 @@ static char **handle_switch_I(char *arg,return next; } -static char **handle_switch_i (char *arg,char **next) +static void add_cmdline_include (char *filename) { -if (*next && !strcmp (arg,"include" )) { -char *name = *++next; -int fd =open (name,O_RDONLY ); - - include_fd = fd; - include = name; -if (fd <0 ) -perror (name); - } -if (*next && !strcmp (arg,"imacros" )) { -char *name = *++next; -int fd =open (name,O_RDONLY ); +int fd =open (filename,O_RDONLY ); +if (fd <0 ) { +perror (filename); +return ; + } +if (cmdline_include_nr >=CMDLINE_INCLUDE ) +die ("too many include files for %s\n" , filename); + cmdline_include[cmdline_include_nr].filename = filename; + cmdline_include[cmdline_include_nr].fd = fd; + cmdline_include_nr++; +} - include_fd = fd; - include = name; -if (fd <0 ) -perror (name); - } +static char **handle_switch_i (char *arg,char **next) +{ +if (*next && !strcmp (arg,"include" )) +add_cmdline_include (*++next); +else if (*next && !strcmp (arg,"imacros" )) +add_cmdline_include (*++next);else if (*next && !strcmp (arg,"isystem" )) {char *path = *++next;if (!path)
@@ -623,12 +626,14 @@ static struct symbol_list *sparse_file(cstatic struct symbol_list *sparse_initial (void ) {struct token *token; +int i;// Prepend any "include" file to the stream. // We're in global scope, it will affect all files! token =NULL ; -if (include_fd >=0 ) - token =tokenize (include, include_fd,NULL , includepath); +for (i = cmdline_include_nr -1 ; i >=0 ; i--) + token =tokenize (cmdline_include[i].filename, cmdline_include[i].fd, + token, includepath);// Prepend the initial built-in stream token =tokenize_buffer (pre_buffer, pre_buffer_size, token);
Index: sparse/token.h
===================================================================
lib.h +9 -0
@@ -33,6 +33,15 @@ struct position { noexpand:1 ; }; +struct cmdline_include { +char *filename; +int fd; +}; + +extern struct cmdline_include cmdline_include[]; +extern int cmdline_include_nr; + +struct ident ;struct token ;struct symbol ;
Hello!
Quoted 9 lines · Christopher Li
Quoted 4 lines · Pavel Roskin
Quoted 3 lines · Pavel Roskin
It seems to me that the existing
add_pre_buffer()mechanism can be used instead. I'm just a bit worried why it wasn't done like this in the first place.I believe the reason it is not in the pre buffer is that it should first search the current directory instead of the source file directory. The command line -include has some subtle differences with #include "filename"
Then maybe we need some variation of #include, e.g. #include_cmdline. And while at that, I think -imacros should be processed slightly differently.
Quoted 8 lines
Quoted 4 lines
There was a reason to worry. Now
create_builtin_stream()is run after the includes have been processed, so that e.g. the Linuxcompiler.htells me that my compiler is too old (because it was included from the command line before__GNUC__was defined).That is the other reason as well :-)
Can you please try this patch and see if it works for you?
I have tried it in current (svn) MadWifi, and the output is much more agreeable, so I think the patch is working.
However, I would prefer that we don't use fixed size arrays. It would be much better in the long term to use the existing "code generator". It would be a more uniform and scalable approach. Besides, the generated code could be dumped for debugging purposes.
By the way, the current MadWifi is a treasure trove for anyone looking to improve sparse. My impression is that most if not all reported problems are bogus.
Signature · 2 lines
Regards, Pavel Roskin