This is a follow up to Bug 764683. Currently, none of the calls to fopen in the standalone updater gracefully handle EINTR signals errors. The general pattern suggested for this is: do { file = fopen("path", "flags"); } while (file == NULL && errno == EINTR);

If we want to support EINTR signals, don't we have to also check them on read/write calls? Do you expect these to happen during an update?

Yes. This *should not* happen during normal operation. However, this code needs to be bullet proof. bionic is the worst PoS libc you could ever imagine, so I mostly want to guard against bugs in there.

Hi > This *should not* happen during normal operation. However, this code needs > to be bullet proof. bionic is the worst PoS libc you could ever imagine, so > I mostly want to guard against bugs in there. Is there an official statement whether bionic automatically restarts interrupted syscall (e.g., like glibc does), or not? I ask, because I thought about going through _all_ sources and adding EINTR handling to _all_ affected calls. I wrote a small test program for triggering an EINTR, but it never did; so I though that bionic might handle this automatically. And there is still the case where you read/write only half of the data you requested...

Even if bionic attempted to implement auto-restart, I wouldn't trust it to get that right :(.

(In reply to Chris Jones [:cjones] [:warhammer] from comment #4) > Even if bionic attempted to implement auto-restart, I wouldn't trust it to > get that right :(. Ok. I opened https://bugzilla.mozilla.org/show_bug.cgi?id=804959 to track this problem, and will work on it with low priority.

B2G no longer supported