git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH v2 02/11] clean: do not pass strbuf by value

From
Jeff King <peff@peff.net>
Date
Aug 2, 2025, 08:38 UTC
Message-ID
<20250802083846.GC3711639@coredump.intra.peff.net>
In-Reply-To
<20250731225433.4028872-3-gitster@pobox.com>
On Thu, Jul 31, 2025 at 03:54:24PM -0700, Junio C Hamano wrote:
Show 6 quoted lines
> This is a catastrophe waiting to happen.  If the callee causes the
> strbuf to be reallocated, the buf[] the caller has will become
> dangling, and when the caller does strbuf_release(), it would result
> in double-free.
> 
> Stop calling the function with misleading call-by-value with strbuf.

This is definitely an improvement, though I wonder if we could go further.

When I saw the original version of this patch, my first thought was: if we are passing a strbuf that is not modified, then why not just pass a const string pointer? It is more flexible for the caller and makes our intention more clear. (A careful reader will note that this does not pass in the length, so it is different if we expect embedded NULs, but I don't think that is the case here).

But it does not quite work in this case because of this:
Show 13 quoted lines
>  	if (is_single) {
> -		choice_list = strbuf_split_max(&input, '\n', 0);
> +		choice_list = strbuf_split_max(input, '\n', 0);
>  	} else {
> -		char *p = input.buf;
> +		char *p = input->buf;
>  		do {
>  			if (*p == ',')
>  				*p = ' ';
>  		} while (*p++);
> -		choice_list = strbuf_split_max(&input, ' ', 0);
> +		choice_list = strbuf_split_max(input, ' ', 0);
>  	}

We do modify the string! However, that code goes away in your next patch when we are able to switch to a split function that handles multiple delimiters. You do still do an in-place split, but isn't strictly needed.

If we swapped patches 2 and 3 in this series, then the strbuf cleanup can just become:

diff --git a/builtin/clean.c b/builtin/clean.c
index a01d130f62..e4b075df70 100644
--- a/builtin/clean.c
+++ b/builtin/clean.c
@@ -477,17 +477,17 @@ static int find_unique(const char *choice, struct menu_stuff *menu_stuff)
  */
 static int parse_choice(struct menu_stuff *menu_stuff,
 			int is_single,
-			struct strbuf input,
+			const char *input,
 			int **chosen)
 {
-	struct string_list choice = STRING_LIST_INIT_NODUP;
+	struct string_list choice = STRING_LIST_INIT_DUP;
 	struct string_list_item *item;
 	int nr = 0;
 	int i;
 
-	string_list_split_in_place_f(&choice, input.buf,
-				     is_single ? "\n" : ", ", -1,
-				     STRING_LIST_SPLIT_TRIM);
+	string_list_split_f(&choice, input,
+			    is_single ? "\n" : ", ", -1,
+			    STRING_LIST_SPLIT_TRIM);
 
 	for_each_string_list_item(item, &choice) {
 		const char *string;
@@ -626,7 +626,7 @@ static int *list_and_choose(struct menu_opts *opts, struct menu_stuff *stuff)
 
 		nr = parse_choice(stuff,
 				  opts->flags & MENU_OPTS_SINGLETON,
-				  choice,
+				  choice.buf,
 				  &chosen);
 
 		if (opts->flags & MENU_OPTS_SINGLETON) {

I dunno. Maybe it is nitpicking, but I think "don't take a strbuf if you
only need a string" is a good general rule. Of course there is only one
caller here, so flexibility is probably not that important.

-Peff
Previous: Junio C HamanoNext: Junio C Hamano
Message 18 of 33 in “do not overuse strbuf_split*()”
  1. 0/9 do not overuse strbuf_split*()Junio C Hamano, Jul 31, 2025
  2. 1/9 wt-status: avoid strbuf_split*()Junio C Hamano, Jul 31, 2025
  3. 2/9 clean: do not pass strbuf by valueJunio C Hamano, Jul 31, 2025
  4. 3/9 clean: do not use strbuf_split*() [part 1]Junio C Hamano, Jul 31, 2025
  5. 4/9 clean: do not use strbuf_split*() [part 2]Junio C Hamano, Jul 31, 2025
  6. 5/9 merge-tree: do not use strbuf_split*()Junio C Hamano, Jul 31, 2025
  7. 6/9 notes: do not use strbuf_split*()Junio C Hamano, Jul 31, 2025
  8. Eric SunshineJul 31, 2025
  9. 7/9 config: do not use strbuf_split()Junio C Hamano, Jul 31, 2025
  10. Eric SunshineJul 31, 2025
  11. 8/9 environment: do not use strbuf_split*()Junio C Hamano, Jul 31, 2025
  12. 9/9 sub-process: do not use strbuf_split*()Junio C Hamano, Jul 31, 2025
  13. Christian CouderJul 31, 2025
  14. Junio C HamanoJul 31, 2025
  15. 00/11 do not overuse strbuf_split*()Junio C Hamano, Jul 31, 2025
  16. 01/11 wt-status: avoid strbuf_split*()Junio C Hamano, Jul 31, 2025
  17. 02/11 clean: do not pass strbuf by valueJunio C Hamano, Jul 31, 2025
  18. Jeff KingAug 2, 2025
  19. Junio C HamanoAug 2, 2025
  20. Jeff KingAug 2, 2025
  21. 03/11 clean: do not use strbuf_split*() [part 1]Junio C Hamano, Jul 31, 2025
  22. 04/11 clean: do not use strbuf_split*() [part 2]Junio C Hamano, Jul 31, 2025
  23. 05/11 merge-tree: do not use strbuf_split*()Junio C Hamano, Jul 31, 2025
  24. Jeff KingAug 2, 2025
  25. 06/11 notes: do not use strbuf_split*()Junio C Hamano, Jul 31, 2025
  26. 07/11 config: do not use strbuf_split()Junio C Hamano, Jul 31, 2025
  27. 08/11 environment: do not use strbuf_split*()Junio C Hamano, Jul 31, 2025
  28. 09/11 sub-process: do not use strbuf_split*()Junio C Hamano, Jul 31, 2025
  29. 10/11 trace2: trim_trailing_newline followed by trim is a no-opJunio C Hamano, Jul 31, 2025
  30. 11/11 trace2: do not use strbuf_split*()Junio C Hamano, Jul 31, 2025
  31. Jeff KingAug 2, 2025
  32. Junio C HamanoAug 2, 2025
  33. Jeff KingAug 2, 2025

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.