Testsuite diffs can be misleading when normalisers are in use
When output fails to match, GHC will output the diff based on the original outputs. However, this is misleading: some "differences" will actually have been normalised by the whitespace and more substantive normalisers we apply. So you really have to compare the normalised output to see what the /real/ difference is.
Unfortunately, the real normalised output is missing newlines, so it's not really suitable for diffing. So it's not really clear what an improved algorithm would be.
This seems to not actually be a problem in practice, since SOP for messages which need heavy normalizing is that when the new message looks good, it's just 'make accept'ed. Can be a little confusing when you're developing normalisers.
Trac metadata
| Trac field | Value |
|---|---|
| Version | 7.11 |
| Type | Task |
| TypeOfFailure | OtherFailure |
| Priority | low |
| Resolution | Unresolved |
| Component | Test Suite |
| Test case | |
| Differential revisions | |
| BlockedBy | |
| Related | |
| Blocking | |
| CC | |
| Operating system | |
| Architecture |