Repository navigation
GH-54643: [C++] Reject impossible dates and a zero day in the strptime timestamp parser - #54657
Conversation
|
Thanks for opening a pull request! This pull request has been automatically closed because you currently have 4 open pull requests, which is more than the limit of 3. Due to the increase in pull requests opened by AI bots, and in order to keep the review queue manageable, Apache Arrow limits contributors without repository access to at most 3 concurrently open pull requests. This helps make sure each pull request gets the attention it needs and that work in progress does not go stale. Once one of your other open pull requests has been merged or closed, you are welcome to reopen this one. See also: |
|
|
…trptime timestamp parser strptime checks the day of month against 31 at most, and some implementations (macOS, the vendored musl used on Windows) accept day 00. The parser rolled impossible dates such as February 30 over into the next month and took a zero day as the first of the month. Initialize the day to the first before parsing, so that a format without a day directive still gives the first of the month, and reject the result if it is not a valid date, as the ISO-8601 parser does. Drafted by Claude Code (an AI agent) and reviewed & approved by pearu.
b46d111 to
7ac2429
Compare
| "2/29/2001 12:34:56", | ||
| // A zero day is rejected rather than | ||
| // taken as the first of the month | ||
| "2/00/2000 12:34:56"}; |
There was a problem hiding this comment.
Could you also add a zero month case?
BTW, could you also apply this change?
diff --git a/cpp/src/arrow/vendored/musl/strptime.c b/cpp/src/arrow/vendored/musl/strptime.c
index eb325108b8..13c0f3172d 100644
--- a/cpp/src/arrow/vendored/musl/strptime.c
+++ b/cpp/src/arrow/vendored/musl/strptime.c
@@ -212,7 +212,7 @@ char *strptime(const char *__restrict s, const char *__restrict f, struct tm *__
*dest = 0;
for (i=1; i<=min+range && isdigit(*s); i*=10)
*dest = *dest * 10 + *s++ - '0';
- if (*dest - min >= range) return 0;
+ if ((unsigned)(*dest - min) >= (unsigned)range) return 0;
*dest -= adj;
switch((char *)dest - (char *)tm) {
case offsetof(struct tm, tm_yday):The upstream has (unsigned)range here: https://git.musl-libc.org/cgit/musl/plain/src/time/strptime.c
Our vendored version doesn't have it. It may have been removed to avoid a -Wsign-compare warning. So I also cast *dest - min to unsigned to avoid the warning.
The change in this PR fixes the problem caused by this difference. So this change isn't strictly needed. But I want to keep our vendored version as close to the upstream as possible.
There was a problem hiding this comment.
Added 0/15/2000 12:34:56: with the vendored strptime on main it became 1999-12-15. Also applied your change; with it the vendored strptime rejects day and month 00 itself, as glibc does.
github-actions
Bot
added
awaiting merge
Drafted by Claude Code (an AI agent) and reviewed & approved by pearu.
|
Thanks. |
Rationale for this change
strptimechecks the day of month against 31 at most, and some implementations (macOS, the vendored musl used on Windows) accept day00, and the vendored musl also month00. The parser rolled an impossible date such as February 30 over into the next month, took a zero day as the first of the month, and a zero month as December of the previous year.What changes are included in this PR?
The parser rejects the parsed date if it is not valid, as the ISO-8601 parser does. To tell a zero day from an omitted one, the day is initialized to the first before parsing, instead of being raised to the first afterwards. The two fixes are in one PR because the zero-day fix relies on the date check. The vendored musl
strptimealso gets back the unsigned range check that upstream musl has, so it rejects day and month00itself.Are these changes tested?
Yes, in
TimestampParser.StrptimeParser(impossible dates, a zero day, a zero month, and February 29 of a leap year) andTimestampParser.StrptimeWithoutDay(a format without a day directive still gives the first of the month). The zero day and zero month cases only fail without the fix on platforms whosestrptimeaccepts them.Are there any user-facing changes?
Impossible dates, a zero day and a zero month no longer parse with
strptimeformats.AI usage disclosure
Developed with AI assistance (Claude Code): the fix, the tests and this description were AI-generated under my direction and reviewed line by line by me. I own and can debug these changes. No copyrighted material is included.