reduce ndslice template bloat #4823
Conversation
|
Do you have any data for the compile time speed difference? |
|
No |
|
Compilation speed for unittests is the same. But this is not a goal. Goal is more clean API, less function names in object files and in the documentation. A lot of prototypes was removed, but API functionality remains the same. |
| @@ -513,6 +477,8 @@ private bool isValidPartialPermutationImpl(size_t N)(in size_t[] perm, ref int[N | |||
| return true; | |||
| } | |||
|
|
|||
| enum toSize_t(size_t i) = i; | |||
| enum isSize_t(alias i) = is(typeof(i) == size_t); | |||
There was a problem hiding this comment.
Can you make all of these private or package?
There was a problem hiding this comment.
std/experimental/ndslice/internal.d is public because it is required to remove existing ndslice modules from Mir but continue add new ndslice modules. So, it should be shared/public for now.
| @@ -329,7 +329,7 @@ Returns: | |||
| n-dimensional slice of the same type | |||
| See_also: $(LREF swapped), $(LREF transposed) | |||
| +/ | |||
| Slice!(N, Range) everted(size_t N, Range)(auto ref Slice!(N, Range) slice) | |||
| Slice!(N, Range) everted(size_t N, Range)(Slice!(N, Range) slice) | |||
There was a problem hiding this comment.
wait, this is a typo, should be reverted
There was a problem hiding this comment.
Evert is a word, though 9il might have meant reverted,
Eversion (from the verb "evert") is the process of turning inside-out.
|
Auto-merge toggled on |
There are 2 commits with different template bloat optimizations.
auto refwas also removed.This PR follows suggestions from @aG0aep6G and @timotheecour from the thread http://forum.dlang.org/thread/njcoyysqeqhbejxperyv@forum.dlang.org?page=1