~bzr-pqm/bzr/bzr.dev : changes

~bzr-pqm/bzr/bzr.dev » Changes from revision 6133.4.12

Rev	Summary	Authors	Date
6133.4.12	Don't traceback for timeouts, and give better documentation ... Don't traceback for timeouts, and give better documentation about why we don't just select.	John Arbash Meinel	13 years ago
6133.4.11	It turns out that if we don't explicitly close the socket, i... It turns out that if we don't explicitly close the socket, it hangs around somewhere. Which means that the client doesn't know that it has been disconnected.	John Arbash Meinel	13 years ago
6133.4.10	Handle the win32 not-a-socket errors. I think this is ~ready... Handle the win32 not-a-socket errors. I think this is ~ready. At least for feedback.	John Arbash Meinel	13 years ago
6133.4.9	Don't forget to actually raise exceptions that aren't handle... Don't forget to actually raise exceptions that aren't handled.	John Arbash Meinel	13 years ago
6133.4.8	Refactor a bit. Use a common _wait_for_descriptor code. Refactor a bit. Use a common _wait_for_descriptor code. Changing the loop to use a 'tight' fast-timeout and a 'slow' poll made it succeed reliably during the test suite. We seem to be getting socket.error(EBADF) for about 1-in-3 tests. My guess is that it is a race condition when you have a long timeout in select. If the handle is bad when select() starts, then you immediately get EBADF. If it is closed on the client, then the server notices and returns that it can be read without blocking. However, if it is closed on the server by another thread, after select has started sleeping, select doesn't end up seeing the status change, and thus has to wait until timeout before you can then notice the file descriptor doesn't mean anything anymore. Oddly, about 1 in 1000 tests fail with select.error rather than socket.error. Even worse, select.error doesn't have an errno attribute (in py 2.7.1), but only a error.args[0] value. By doing a fast-select (and setting the timeout really low in the test suite), tests clean up reliably and don't hang for > second. The long poll allows the test suite to have hiccups/gc/etc without failing a test. The 4.0s timeout during the test suite is because we have a hard 5.0s 'thread has hung' timeout. 100ms fast timeout lets us detect shutdown races, and clean up without waiting for the hard-timeout. Now to test on Windows, and figure out the configuration story.	John Arbash Meinel	13 years ago
6133.4.7	Currently causes 5-10 failures in bt.per_branch.test_branch ... Currently causes 5-10 failures in bt.per_branch.test_branch Remote out of 150 tests. Adding a print, statement after checking the return value from rs, but before returning from the _wait code, fixes all failures... :(	John Arbash Meinel	13 years ago
6133.4.6	Some testing shows that if the client closes its socket, the... Some testing shows that if the client closes its socket, the server immediately returns.	John Arbash Meinel	13 years ago
6133.4.5	This new 'check' causes issues with at least the blackbox te... This new 'check' causes issues with at least the blackbox tests. Hung threads mostly, but if you set the timeout low, it can cause timeouts, too. Needs more investigation.	John Arbash Meinel	13 years ago
6133.4.4	Create a default timeout of 5 minutes. Create a default timeout of 5 minutes. Have the test suite override the default timeout down to 10 seconds. Tests should never actually wait for data, it is just a back-stop in case there is a bug, so that we don't have a test hang for 5 minutes waiting for it to timeout. 10s is a bit long, but since earlier testing showed that sometimes we'll spike to about 1+s for what should otherwise be a fast test, we want a bit of margin to avoid spurious failures.	John Arbash Meinel	13 years ago
6133.4.3	use some 'with' statements for better cleanups. use some 'with' statements for better cleanups.	John Arbash Meinel	13 years ago
6133.4.2	Implement waiting for Pipe mediums, including handling ones Implement waiting for Pipe mediums, including handling ones that we cannot select on.	John Arbash Meinel	13 years ago
6133.4.1	Add a _wait_for_bytes_with_timeout helper function. Add a _wait_for_bytes_with_timeout helper function. This adds a select.select() call, indicating whether there are any bytes on the socket that we can read, or whether we timeout waiting for more content.	John Arbash Meinel	13 years ago
6133	(jameinel) Estimate bytes-to-return more accurately for (jameinel) Estimate bytes-to-return more accurately for Repository.get_parent_map, allowing closer to 64kB to be returned per query. (John A Meinel)	Patch Queue Manager	13 years ago
6132	(jr) Do not use i18n for tests (Jonathan Riddell) (jr) Do not use i18n for tests (Jonathan Riddell)	Patch Queue Manager	13 years ago
6131	(jr) Use gettext.NullTranslations in i18n to allow use of i1... (jr) Use gettext.NullTranslations in i18n to allow use of i18n even when translations are not turned on. (Jonathan Riddell)	Patch Queue Manager	13 years ago
6130	(jelmer) More fixes for the interface tests when run against... (jelmer) More fixes for the interface tests when run against foreign tests. (Jelmer Vernooij)	Patch Queue Manager	13 years ago
6129	(jelmer) Misc fixes for running the regular bzr tests agains... (jelmer) Misc fixes for running the regular bzr tests against foreign branches. (Jelmer Vernooij)	Patch Queue Manager	13 years ago
6128	(vila) Properly decode command lines with embedded quotes fo... (vila) Properly decode command lines with embedded quotes for external merge tools. (Gordon Tyler)	Patch Queue Manager	13 years ago
6127	(jelmer) Fix some more tests when run against foreign bzr pl... (jelmer) Fix some more tests when run against foreign bzr plugins. (Jelmer Vernooij)	Patch Queue Manager	13 years ago
6126	(vila) Document fix for #842993 in release notes. (Martin vo... (vila) Document fix for #842993 in release notes. (Martin von Gagern) (Martin von Gagern)	Patch Queue Manager	13 years ago

Older »