strptime problems

strptime problems

[Rich Felker]

It was raised somewhere (on IRC, I think) that strptime still has some
failure paths where functionality is unimplemented. These should have
been caught and fixed a long time ago, but they weren't, so I'd like
to get it done now. However I'm not clear on what the behavior of
strptime should be when reading "derived" fields that don't actually
correspond to anything in the "struct tm". Should it just parse and
ignore them, or should it somehow convert back? The week-based-year
stuff is probably the biggest question but there are others too.

Rich

Re: strptime problems

[Isaac Dunham]

On Thu, May 08, 2014 at 12:32:14AM -0400, Rich Felker wrote:
> It was raised somewhere (on IRC, I think) that strptime still has some
> failure paths where functionality is unimplemented. These should have
> been caught and fixed a long time ago, but they weren't, so I'd like
> to get it done now. However I'm not clear on what the behavior of
> strptime should be when reading "derived" fields that don't actually
> correspond to anything in the "struct tm". Should it just parse and
> ignore them, or should it somehow convert back? The week-based-year
> stuff is probably the biggest question but there are others too.

I'd say it shouuld convert back.
"%U %w %Y" is enough information to find the day referred to.

HTH,
Isaac Dunham

Re: strptime problems

[Szabolcs Nagy]

* Rich Felker <dalias@libc.org> [2014-05-08 00:32:14 -0400]:
> to get it done now. However I'm not clear on what the behavior of
> strptime should be when reading "derived" fields that don't actually
> correspond to anything in the "struct tm". Should it just parse and
> ignore them, or should it somehow convert back? The week-based-year
> stuff is probably the biggest question but there are others too.

strptime seems to be underspecified, it's not clear what happens
in case of inconsistent input, what tm_* fields should be filled
out, etc

glibc sets tm.tm_wday when it can be calculated, musl doesn't

glibc sets __tm_gmtoff so strptime depends on the current tz setting

glibc supports %s in strptime (non-posix), musl doesn't

in musl invalid format strings don't fail now

in the numeric_range logic i think a <= can be changed into <

--- a/src/time/strptime.c
+++ b/src/time/strptime.c
@@ -138,10 +138,12 @@ char *strptime(const char *restrict s, const char *restrict f, struct tm *restri
                case '%':
                        if (*s++ != '%') return 0;
                        break;
+               default:
+                       return 0;
                numeric_range:
                        if (!isdigit(*s)) return 0;
                        *dest = 0;
-                       for (i=1; i<=min+range && isdigit(*s); i*=10)
+                       for (i=1; i<min+range && isdigit(*s); i*=10)
                                *dest = *dest * 10 + *s++ - '0';
                        if (*dest - min >= (unsigned)range) return 0;
                        *dest -= adj;