Re: [9front] libstdio fscanf regression

[ori@eigenstate.org]

> Hello,
> 
> Commit 7551 (fix '%[]' specifiers and '%n' (thanks phil9)) seems to have
> broken libstdio's *scanf (ape version untested).
> 
> Test case is games/doom's configuration file loading, which uses fscanf:
> /sys/src/games/doom/m_misc.c:314
> 
> ...
> 
> 	    if (fscanf (f, "%79s %[^\n]\n", def, strparm) == 2)
> 
> ...

Thanks for the report -- I think this is what we want. At least,
it works for your test case, and it tracks with what I'd expect
the semantics to be. I haven't pored over the spec yet.

Test code:

#include <u.h>
#include <libc.h>
#include <stdio.h>

void
main(int argc, char **argv)
{
	char def[80], strparm[128];

	print("%d\n", scanf("%79s %[^\n]\n", def, strparm));
	exits(nil);
}

Patch:

diff -r 35459627f401 sys/src/ape/lib/ap/stdio/vfscanf.c
--- a/sys/src/ape/lib/ap/stdio/vfscanf.c	Wed May 13 18:50:01 2020 -0700
+++ b/sys/src/ape/lib/ap/stdio/vfscanf.c	Thu May 14 17:48:32 2020 -0700
@@ -416,7 +416,7 @@
 		}
 		if(!match(c, pat)){
 			nungetc(c, f);
-			return 0;
+			return nn > 0;
 		}
 		if(store)
 			*s++=c;
diff -r 35459627f401 sys/src/libstdio/vfscanf.c
--- a/sys/src/libstdio/vfscanf.c	Wed May 13 18:50:01 2020 -0700
+++ b/sys/src/libstdio/vfscanf.c	Thu May 14 17:48:32 2020 -0700
@@ -339,6 +339,7 @@
 	int c, nn;
 	register char *s;
 	register const char *pat;
+
 	pat=++fmtp;
 	if(*fmtp=='^') fmtp++;
 	if(*fmtp!='\0') fmtp++;
@@ -354,7 +355,7 @@
 		}
 		if(!match(c, pat)){
 			nungetc(c, f);
-			return 0;
+			return nn > 0;
 		}
 		if(store) *s++=c;
 		nn++;