Invalid file on bsdtar command line results in internal errors...

[kientzle]

This was reported to the FreeBSD bugzilla as:

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=213092

# tar cvf - foo that this 404 that this 404 that this > /dev/null
a foo
a foo/bar
a that
a this
tar: 404: Cannot stat: No such file or directory
tar: INTERNAL ERROR: Function 'archive_read_disk_open' invoked with archive structure in state 'header', should be in state 'new/closed': Unknown error: -1
tar: INTERNAL ERROR: Function 'archive_read_disk_open' invoked with archive structure in state 'header', should be in state 'new/closed': Unknown error: -1
tar: INTERNAL ERROR: Function 'archive_read_disk_open' invoked with archive structure in state 'header', should be in state 'new/closed': Unknown error: -1
tar: INTERNAL ERROR: Function 'archive_read_disk_open' invoked with archive structure in state 'header', should be in state 'new/closed': Unknown error: -1
tar: INTERNAL ERROR: Function 'archive_read_disk_open' invoked with archive structure in state 'header', should be in state 'new/closed': Unknown error: -1
tar: Error exit delayed from previous errors.
# echo $?
1

It at least returns 1, but it doesn't stop execution. It seems to get into a pretty yucky state. I think maybe tar isn't handling the return from libarchive properly? I think tar should quit immediately on a failure like this, rather than spinning its wheels in a weird state.

In order to see the libarchive error, you have to give tar at least one more file argument after a file that doesn't exist.

Thank you!

[mmatuska]

I can confirm this behavior. Simply use:
tar cf new.tar notexisting1 notexisting2

I will look into it later.

[mmatuska]

I don't understand why this was unnoticed and not catched by any test.

[ghost]

Thank you!

How hard would it be write a test case for this? Have you done that in a follow up commit?

[mmatuska]

Hi, I didn't do that yet. This will probably require changing the test framework (adding new functions). The only thing we can do is look for "INTERNAL ERROR" in the stderr output. I think we should go a different way and introduce additional exit codes for tar.

[kientzle]

Adding exit codes for tar is tricky: I could support having a special exit code for "INTERNAL ERROR" except that libarchive doesn't return anything special for "INTERNAL ERROR" (other than ARCHIVE_FATAL). And adding new error codes to libarchive is very hard at this point. (A significant but probably worthwhile project would add a more detailed error code to libarchive so that callers can get precise machine-readable error info.)

However, it's pretty clear that tar should never, ever print "INTERNAL ERROR", so it might be worth adding a test function that takes a file name (presumably of captured stderr) and fails if that string appears anywhere in the file. Something like:

static void
assertCleanStderr(const char *pathname) {
   ... read file, check for 'INTERNAL ERROR' (and anything else that should never appear in stderr) ...
}

Then we could start augmenting various tests with calls to assertCleanStderr(errOut) to look for recurrence of this issue in other places.