TestCase.flushWarnings shouldn't use source files to implement its "offendingFunctions" feature

[exarkun]

Using source files to determine which function a warning was emitted from is unreliable. It fails if:

1. The source file is missing
2. The source file is out of date with the bytecode file
3. It's kind of slow

dis.findlinestarts can provide the necessary information and is now public.

[exarkun]

(In [25781]) Branching to 'findlinestarts-3598'

[exarkun]

(In [25782]) use dis.findlinestarts instead of inspect.getsourcelines

refs #3598

[exarkun]

Think this is in good shape now.

[radix]

[1] Need to mention copyright of that code copied from Python

[2] in test_renamedSource:

• a: you should not call the *original* filename "renamedsourcefile", since that is confusing.
• b: I don't think this comment is correct:

+ # Drop it from memory so we can import it again later

You don't actually need to drop it so you can import it again, since it's going to be imported with the different name, right? Of course, you do need to clean it up, but you can do that at the end.

[3] Please change that function to not use single-letter variables that are used across more than one line.

[4] I don't know whether the following expression has off-by-one errors:

+ if not (first <= w.lineno <= last):

Can you please add tests for all of the edge cases, thank you.

[exarkun]

(In [25846]) test edge cases for line number stuff

refs #3598

[exarkun]

(In [25847]) rename single letter variable "w" to something more descriptive

refs #3598

[exarkun]

(In [25848]) add/adjust comments in new test method to be more descriptive/accurate

refs #3598

[exarkun]

(In [25849]) document copyright holder and license of _findlinestarts

refs #3598

[exarkun]

All points addressed, I think. Build results: http://buildbot.twistedmatrix.com/boxes-supported?branch=/branches/findlinestarts-3598

[exarkun]

This is a small part of #3554.

[mwh]

I think it looks fine. I have two small comments.

1. Saying "inspect.getabsfile(aFunction) returns bad results sometimes." is a bit useless. What are bad results? When does it generate them?

2. I know you apologise for it, but 'renamedsourcefile.py' is never actually renamed. I think something plain 'module.py' would be clearer.

These are minor enough that I don't want to review this again :) Please fix and land.

[exarkun]

(In [26041]) explain the getabsfile problem in more detail; change the name of the module used in the renamed-source-file-test to be more clear

refs #3598

[exarkun]

(In [26043]) Merge findlinestarts-3598

Author: exarkun Reviewer: radix, mwhudson Fixes: #3598

Change the implementation of twisted.trial.unittest.TestCase.flushWarnings so that it works properly when a warning is emitted from a module which is defined by a source file which has been renamed since it was compiled. Previously, the rename would cause file names to disagree in certain places and prevent warnings emitted from code in a renamed file from being flushed.