Added support for the rest of the char types #192
Conversation
Added support for the other char types
include/fast_float/ascii_number.h
Outdated
| fastfloat_really_inline FASTFLOAT_CONSTEXPR20 | ||
| uint32_t parse_eight_digits_unrolled(const char *chars) noexcept { | ||
| uint32_t parse_eight_digits_unrolled(TCH const * chars) noexcept { |
There was a problem hiding this comment.
This routine expects ASCII/UTF-8. It will fail on UTF-8 inputs, for example. (Update: I was wrong.)
There was a problem hiding this comment.
read_u64 handles the conversion of integer size between char types
There was a problem hiding this comment.
I stand corrected. However, it would still be be needed to check that it makes sense to do so: it is a very cheap check given ASCII inputs, but if you need to do more work, it may end up being a net negative. So we would need to benchmark with and without this optimization to make sure it is warranted.
There was a problem hiding this comment.
I understand, my reasoning for all of this is that I needed an unicode version, so I don't have to double buffer with fast or wcstod, now I'm going to add hex support using radix bases, and finally an ignore list, as for the tests I'll update them as I go, I guess I'll add a new enum 'infer' to chars_format to work as a pseudo drop-in replacement for wcstod behaviour.
There was a problem hiding this comment.
Sure. Great work. What you are doing is perfectly reasonable.
Ignore my initial negative and misguided comment.
There was a problem hiding this comment.
No worries, it's such a delicate change, and everywhere.
There was a problem hiding this comment.
I stand corrected. However, it would still be be needed to check that it makes sense to do so: it is a very cheap check given ASCII inputs, but if you need to do more work, it may end up being a net negative. So we would need to benchmark with and without this optimization to make sure it is warranted.
I have a version of fast_float in which I've attempted to optimize for other char_types using opt-in SIMD. It also adds support for different parsing rules (JSON parsing rules, opt-in inf/nan etc.) --> if this is okay I'm happy to submit a PR or modify this one.
|
Thanks. We are going to need tests for all these types before we can consider this. |
|
Note that it looks highly valuable, so if we can manage to have decent testing, then we should definitively merge this. |
|
@Pharago There is no harm in adding your name to the contributor list as part of this PR. |
Removed storage class declarations from explicit template specializations of string constants
|
Running tests. |
|
The Ubuntu 18.04 failure is probably because Ubuntu 18.04 is no longer supported in CI. I need to remove it. |
|
@Pharago We are going to need tests for this. We don't need to reproduce all tests (it would take a long time) but at least some of the tests should be extended to include different char types. |
No description provided.