fix bug 65667

[pilif]

This PR is a fix for bug 65667

the idea behind ftp_nb_get is for it to be followed by multiple calls
to ftp_nb_continue in order to download a file piece-by-piece.

As such, it's unwise to close the stream used to write the downloaded
data to when the file hasn't been completely downloaded within the first
call to ftp_nb_get.

This regression was added in a93a462
and this patch restores the behavior that was seen pre-patch.

[pilif]

(ext/session/tests/session_set_save_handler_class_005.phpt which failed the build also fails without my patch)

[m6w6]

Yeah, the test suite is not 100% stable on travis yet.

[m6w6]

Looks like a regression
http://git.php.net/?p=php-src.git;a=commit;h=a93a462dcefd62e07963dd2da506fbb3409c88b5

[m6w6]

So, the test suite doesn't fail now and I suspect it won't fail after merging your PR :) i.e. proof is missing.
Mind adding a test? :)

[pilif]

ok. I've added a testcase now.

I've rebased the commits such that the failing test case comes before the commit which fixes the bug.

Aside of that, let me switch into rant-mode for a second:

Here's what I've done for this bug:

• After getting bitten by the issue (caused by a undocumented feature-addition to a point release) I searched the bug database in order to not report a duplicate
• After confirming seeing the bug on my part, I went ahead and investigated the cause
• I fixed the issue (respecting whitespace conventions - contrary to the commit that introduced the bug, btw).
• I went through the history of the file in order to find the commit where the bug was introduced and pointed to that in the PHP bug
• I commented on the offending commit, noting the bug as it was introduced.

After doing all that instead of a thank you and maybe even a merge of the PR, the thing I get is a complaint about a missing test. Yeah. I understand you'd like to see a testcase, but just consider what all went wrong to cause this bug:

• The test-case attached to the commit that added the bug cheats around triggering the bug
• The commit that caused the bug had inconsitent whitespace
• The bug was introduced by changing lines that were never supposed to have been commited: the offending lines are completely unrelated to the commit and the commit message given.
• The commit resulted in a feature-change in a point release which
• was not documented in the changelog

As such, I would say that this commit should not have passed review, but yet it did.

On the other hand, my little fix which unbreaks the, now completely broken ftp_nb_get() gets complained about because of a missing test, despite the fact that there never was a test for ftp_nb_continue() and that it only re-adds code that was there before and likely removed by accident.

Come on.

Anyways. Rant over. As somebody using ftp_nb_get(), I'd naturally like to see this fixed :-)

[m6w6]

Congrats, you've done a great job. Though, you could have reviewed at least 3 other bugs while writing this rant. The "Thank you" usually comes from a template when the bug is closed, but maybe we can get github to post a "Thank you" when requesting a pull?

[php-pulls]

Comment on behalf of nikic at php.net:

Merged in 96cc419.

[nikic]

@pilif I'm sorry that the incorrect change in a93a462 got through code review. We prefer having a test for every change for exactly that reason: The test will prevent the regression from ever being introduced again. Even if the reviewer misses the change, the test will not ;)

[pilif]

@nikic no worries. Thank you so much for applying this and sorry for my rant - I guess I had a bad day today :(

[kaplanlior]

@pilif As long as you keep providing patches, you're welcome to rant (:

[pilif]

If only my C was better - then you'd get plenty of them. Until then I'll have to constrain myself to the low-hanging fruits :-)