Declaration Order
-----------------
-Here is the order in which code should be laid out in a file:
+Here is the preferred order in which code should be laid out in a file:
- commented program name and one-line description
- commented author name and email address(es)
- commented GPL boilerplate
- commented longer description / notes for the program (if needed)
- - #includes and #defines
+ - #includes of .h files with angle brackets (<>) around them
+ - #includes of .h files with quotes ("") around them
+ - #defines (if any, note the section below titled "Avoid the Preprocessor")
- const and global variables
- function declarations (if necessary)
- function implementations
\s\s\s\sif (expr) {\n\tstmt; --ick.) The only exception to this rule is
multi-line comments that use an asterisk at the beginning of each line, i.e.:
- /t/*
- /t * This is a block comment.
- /t * Note that it has multiple lines
- /t * and that the beginning of each line has a tab plus a space
- /t * except for the opening '/*' line where the slash
- /t * is used instead of a space.
- /t */
+ \t/*
+ \t * This is a block comment.
+ \t * Note that it has multiple lines
+ \t * and that the beginning of each line has a tab plus a space
+ \t * except for the opening '/*' line where the slash
+ \t * is used instead of a space.
+ \t */
Furthermore, The preference is that tabs be set to display at four spaces
wide, but the beauty of using only tabs (and not spaces) at the beginning of
Don't do this either:
while (!done){
+
do{
And for heaven's sake, don't do this:
while (!done)
{
+
do
{
Do this instead:
while (!done) {
+
do {
+If you have long logic statements that need to be wrapped, then uncuddling
+the bracket to improve readability is allowed. Generally, this style makes
+it easier for reader to notice that 2nd and following lines are still
+inside 'if':
-Paren Spacing
-~~~~~~~~~~~~~
+ if (some_really_long_checks && some_other_really_long_checks
+ && some_more_really_long_checks
+ && even_more_of_long_checks
+ ) {
+ do_foo_now;
-Put a space between C keywords and left parens, but not between
-function names and the left paren that starts it's parameter list (whether it
-is being declared or called). Examples:
+Spacing around Parentheses
+~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+Put a space between C keywords and left parens, but not between function names
+and the left paren that starts it's parameter list (whether it is being
+declared or called). Examples:
Don't do this:
...
baz = my_func(1, 2);
+Also, don't put a space between the left paren and the first term, nor between
+the last arg and the right paren.
+
+ Don't do this:
+
+ if ( x < 1 )
+ strcmp( thisstr, thatstr )
+
+ Do this instead:
+
+ if (x < 1)
+ strcmp(thisstr, thatstr)
+
Cuddled Elses
~~~~~~~~~~~~~
}
+Labels
+~~~~~~
+
+Labels should start at the beginning of the line, not indented to the block
+level (because they do not "belong" to block scope, only to whole function).
+
+ if (foo) {
+ stmt;
+ label:
+ stmt2;
+ stmt;
+ }
+
+(Putting label at position 1 prevents diff -p from confusing label for function
+name, but it's not a policy of busybox project to enforce such a minor detail).
+
Variable and Function Names
Exceptions:
- - Enums, macros, and constant variables should all be in upper-case with
- words optionally seperatedy by underscores (i.e. FIFOTYPE, ISBLKDEV()).
+ - Enums, macros, and constant variables are occasionally written in all
+ upper-case with words optionally separated by underscores (i.e. FIFO_TYPE,
+ ISBLKDEV()).
- Nobody is going to get mad at you for using 'pvar' as the name of a
variable that is a pointer to 'var'.
-Note: The Busybox codebase is very much a mixture of code gathered from a
-variety of sources. This explains why the current codebase contains such a
-hodge-podge of different naming styles (Java, Pascal, K&R, just-plain-weird,
-etc.). The K&R guideline explained above should therefore be used on new files
-that are added to the repository. Furthermore, the maintainer of an existing
-file that uses alternate naming conventions should -- at his own convenience
--- convert those names over to K&R style; converting variable names is a very
-low priority task. Perhaps in the future we will include some magical Perl
-script that can go through and convert variable names, left as an exercise for
-the reader for now.
-For the time being, if you want to do a search-and-replace of a variable name
-in different files, do the following in the busybox directory:
+Converting to K&R
+~~~~~~~~~~~~~~~~~
+
+The Busybox codebase is very much a mixture of code gathered from a variety of
+sources. This explains why the current codebase contains such a hodge-podge of
+different naming styles (Java, Pascal, K&R, just-plain-weird, etc.). The K&R
+guideline explained above should therefore be used on new files that are added
+to the repository. Furthermore, the maintainer of an existing file that uses
+alternate naming conventions should, at his own convenience, convert those
+names over to K&R style. Converting variable names is a very low priority
+task.
+
+If you want to do a search-and-replace of a single variable name in different
+files, you can do the following in the busybox directory:
$ perl -pi -e 's/\bOldVar\b/new_var/g' *.[ch]
+If you want to convert all the non-K&R vars in your file all at once, follow
+these steps:
+
+ - In the busybox directory type 'examples/mk2knr.pl files-to-convert'. This
+ does not do the actual conversion, rather, it generates a script called
+ 'convertme.pl' that shows what will be converted, giving you a chance to
+ review the changes beforehand.
+
+ - Review the 'convertme.pl' script that gets generated in the busybox
+ directory and remove / edit any of the substitutions in there. Please
+ especially check for false positives (strings that should not be
+ converted).
+
+ - Type './convertme.pl same-files-as-before' to perform the actual
+ conversion.
+
+ - Compile and see if everything still works.
+
+Please be aware of changes that have cascading effects into other files. For
+example, if you're changing the name of something in, say utility.c, you
+should probably run 'examples/mk2knr.pl utility.c' at first, but when you run
+the 'convertme.pl' script you should run it on _all_ files like so:
+'./convertme.pl *.[ch]'.
+
Avoid The Preprocessor
Don't do this:
- #define var 80
+ #define CONST 80
Do this instead, when the variable is in a header file and will be used in
- several source files:
-
- const int var = 80;
+ several source files:
- Or do this when the variable is used only in a single source file:
+ enum { CONST = 80 };
- static const int var = 80;
-
-Declaring variables as '[static] const' gives variables an actual type and
-makes the compiler do type checking for you; the preprocessor does _no_ type
-checking whatsoever, making it much more error prone. Declaring variables with
-'[static] const' also makes debugging programs much easier since the value of
-the variable can be easily queried and displayed.
+Although enum may look ugly to some people, it is better for code size.
+With "const int" compiler may fail to optimize it out and will reserve
+a real storage in rodata for it! (Hopefully, newer gcc will get better
+at it...). With "define", you have slight risk of polluting namespace
+(#define doesn't allow you to redefine the name in the inner scopes),
+and complex "define" are evaluated each time they uesd, not once
+at declarations like enums. Also, the preprocessor does _no_ type checking
+whatsoever, making it much more error prone.
The Folly of Macros
}
Static inline functions are greatly preferred over macros. They provide type
-safety, have no length limitations, no formatting limitations, and under gcc
-they are as cheap as macros. Besides, really long macros with backslashes at
-the end of each line are ugly as sin.
+safety, have no length limitations, no formatting limitations, have an actual
+return value, and under gcc they are as cheap as macros. Besides, really long
+macros with backslashes at the end of each line are ugly as sin.
The Folly of #ifdef
~~~~~~~~~~~~~~~~~~~
Code cluttered with ifdefs is difficult to read and maintain. Don't do it.
-Instead, put your ifdefs in a header, and conditionally define 'static inline'
-functions, (or *maybe* macros), which are used in the code.
+Instead, put your ifdefs at the top of your .c file (or in a header), and
+conditionally define 'static inline' functions, (or *maybe* macros), which are
+used in the code.
Don't do this:
ret = my_func(bar, baz);
if (!ret)
return -1;
- #ifdef BB_FEATURE_FUNKY
+ #ifdef CONFIG_FEATURE_FUNKY
maybe_do_funky_stuff(bar, baz);
#endif
(in .h header file)
- #ifndef BB_FEATURE_FUNKY
- static inline void maybe_do_funky_stuff (int bar, int baz) {}
+ #if ENABLE_FEATURE_FUNKY
+ static inline void maybe_do_funky_stuff(int bar, int baz)
+ {
+ /* lotsa code in here */
+ }
+ #else
+ static inline void maybe_do_funky_stuff(int bar, int baz) {}
#endif
(in the .c source file)
maybe_do_funky_stuff(bar, baz);
The great thing about this approach is that the compiler will optimize away
-the "no-op" case when the feature is turned off.
+the "no-op" case (the empty function) when the feature is turned off.
Note also the use of the word 'maybe' in the function name to indicate
conditional execution.
certain library functions are (mis)used. The following table offers a summary
of some of the more notorious troublemakers:
-function overflows preferred
-----------------------------------------
-strcpy dest string strncpy
-strcat dest string strncat
-gets string it gets fgets
-getwd buf string getcwd
-[v]sprintf str buffer [v]snprintf
-realpath path buffer use with pathconf
-[vf]scanf its arguments just avoid it
+function overflows preferred
+-------------------------------------------------
+strcpy dest string safe_strncpy
+strncpy may fail to 0-terminate dst safe_strncpy
+strcat dest string strncat
+gets string it gets fgets
+getwd buf string getcwd
+[v]sprintf str buffer [v]snprintf
+realpath path buffer use with pathconf
+[vf]scanf its arguments just avoid it
The above is by no means a complete list. Be careful out there.
Avoid Big Static Buffers
------------------------
-First, some background to put this discussion in context: Static buffers look
+First, some background to put this discussion in context: static buffers look
like this in code:
/* in a .c file outside any functions */
- static char *buffer[BUFSIZ]; /* happily used by any function in this file,
+ static char buffer[BUFSIZ]; /* happily used by any function in this file,
but ick! big! */
The problem with these is that any time any busybox app is run, you pay a
memory penalty for this buffer, even if the applet that uses said buffer is
not run. This can be fixed, thusly:
- static char *buffer
+ static char *buffer;
...
other_func()
{
declared on the stack in the *_main() function and made available globally by
assigning them to a global pointer thusly:
- static char *pbuffer
+ static char *pbuffer;
...
other_func()
{
pbuffer = buffer; /* but available globally */
...
-Thus:
- - global static buffers are eliminated
- - we don't grow the text segment as much because no malloc() call is made;
- memory is automatically allocated on the stack when execution context
- enters the function. (We still grow text a little bit because of the
- assignment, but that's cheap compared to a function call.)
- - the buffer is still available globally via the pointer
+This last approach has some advantages (low code size, space not used until
+it's needed), but can be a problem in some low resource machines that have
+very limited stack space (e.g., uCLinux).
+
+A macro is declared in busybox.h that implements compile-time selection
+between xmalloc() and stack creation, so you can code the line in question as
+
+ RESERVE_CONFIG_BUFFER(buffer, BUFSIZ);
+
+and the right thing will happen, based on your configuration.
+
+Another relatively new trick of similar nature is explained
+in keep_data_small.txt.
formatting, options, etc.), model it after the equivalent GNU program.
Doesn't matter how that program behaves on some other flavor of *NIX; doesn't
matter what the POSIX standard says or doesn't say, just model Busybox
-programs after their GNU counterparts and nobody has to get hurt.
+programs after their GNU counterparts and it will make life easier on (nearly)
+everyone.
The only time we deviate from emulating the GNU behavior is when:
would be required, lots more memory would be used, etc.)
- The difference is minor or cosmetic
-A note on the 'cosmetic' case: Output differences might be considered
+A note on the 'cosmetic' case: output differences might be considered
cosmetic, but if the output is significant enough to break other scripts that
use the output, it should really be fixed.
if (foo)
stmt1;
new_line();
- stmt2
+ stmt2;
stmt3;
And the resulting behavior of your program would totally bewilder you. (Don't
Furthermore, you should put a single comment (not necessarily one line, just
one comment) before the block, rather than commenting each and every line.
-There is an optimal ammount of commenting that a program can have; you can
+There is an optimal amount of commenting that a program can have; you can
comment too much as well as too little.
-A picture is really worth a thousand words here, so here is an example that
-illustrates emphasizing logical blocks:
+A picture is really worth a thousand words here, the following example
+illustrates how to emphasize logical blocks:
- while (line = get_line_from_file(fp)) {
+ while (line = xmalloc_fgets(fp)) {
/* eat the newline, if any */
- if (line[strlen(line)-1] == '\n') {
- line[strlen(line)-1] = '\0';
- }
+ chomp(line);
/* ignore blank lines */
if (strlen(file_to_act_on) == 0) {
/* clean up */
free(line);
}
+
+
+Processing Options with getopt
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+If your applet needs to process command-line switches, please use getopt32() to
+do so. Numerous examples can be seen in many of the existing applets, but
+basically it boils down to two things: at the top of the .c file, have this
+line in the midst of your #includes, if you need to parse long options:
+
+ #include <getopt.h>
+
+Then have long options defined:
+
+ static const struct option <applet>_long_options[] = {
+ { "list", 0, NULL, 't' },
+ { "extract", 0, NULL, 'x' },
+ { NULL, 0, NULL, 0 }
+ };
+
+And a code block similar to the following near the top of your applet_main()
+routine:
+
+ char *str_b;
+
+ opt_complementary = "cryptic_string";
+ applet_long_options = <applet>_long_options; /* if you have them */
+ opt = getopt32(argc, argv, "ab:c", &str_b);
+ if (opt & 1) {
+ handle_option_a();
+ }
+ if (opt & 2) {
+ handle_option_b(str_b);
+ }
+ if (opt & 4) {
+ handle_option_c();
+ }
+
+If your applet takes no options (such as 'init'), there should be a line
+somewhere in the file reads:
+
+ /* no options, no getopt */
+
+That way, when people go grepping to see which applets need to be converted to
+use getopt, they won't get false positives.
+
+For more info and examples, examine getopt32.c, tar.c, wget.c etc.