| msg24579 - (view) |
Author: Stefan Behnel (scoder) *  |
Date: 2005-03-14 11:20 |
| The documentation about list.sort() and sorted() does not state how the keyword arguments cmp and key interact. Aparently, if key is supplied, cmp works on the result of the key function and no longer on the item itself. This should be documented. |
|
|
| msg24580 - (view) |
Author: Peter van Kampen (pterk) |
Date: 2005-07-03 20:00 |
| Logged In: YES user_id=174455 I have submittted a doc patch (#1231890) that adds a note that reads: If both key and cmp are supplied the key function is applied before the cmp function. (You can check listsort in listobject.c to verify) |
|
|
| msg24581 - (view) |
Author: Georg Brandl (georg.brandl) * |
Date: 2005-07-03 20:22 |
| Logged In: YES user_id=1188172 Thanks for the report and the patch. Committed as Doc/lib/libstdtypes.tex r1.183, r1.170.2.12. |
|
|
| msg24582 - (view) |
Author: Raymond Hettinger (rhettinger) * |
Date: 2005-07-04 04:13 |
| Logged In: YES user_id=80475 Actually, it should NOT be documented because 1) it is an implementation specific detail and 2) using key and cmp together is almost always the wrong thing to do. |
|
|
| msg24583 - (view) |
Author: Stefan Behnel (scoder) * |
Date: 2005-07-04 09:52 |
| Logged In: YES user_id=313935 ... except, for example, when implementing classes that behave like builtins and stdlib. So, if it's an implementation detail, why not document that and give the hint that the current CPython implementation applies key before cmp? I do not think it matters that using them together is "almost always the wrong thing to do". In that case, it should either become impossible or at least merit an understandable warning in the documentation. |
|
|
| msg24584 - (view) |
Author: Georg Brandl (georg.brandl) * |
Date: 2005-07-04 10:21 |
| Logged In: YES user_id=1188172 But in this case, don't you think that the implementor would in every case look up how it's done in the source code? Raymond, please decide. |
|
|
| msg24585 - (view) |
Author: Raymond Hettinger (rhettinger) * |
Date: 2005-07-04 14:03 |
| Logged In: YES user_id=80475 The docs are not meant to capture all implementation details. Building lookalike objects only need to match the documented, promised behavior -- the rest can be arbitrary. And, yes, it does matter that using the two together is almost always the wrong thing to do. The docs are supposed to point people in the right direction -- it makes Python easier to use. Please revert this patch. |
|
|
| msg24586 - (view) |
Author: Georg Brandl (georg.brandl) * |
Date: 2005-07-04 14:20 |
| Logged In: YES user_id=1188172 Done. Doc/lib/libstdtypes.tex r1.170.2.13, r1.184. |
|
|