Introduce a heuristic to print floating point numbers more coincely #89
Conversation
|
I guess I do need to fix the tests somehow. I can't buld them because meson says |
marzer
left a comment
marzer
left a comment
There was a problem hiding this comment.
Also need to regen toml.hpp ^_^
| const auto res = hexfloat | ||
| ? std::to_chars(buf, buf + sizeof(buf), val, std::chars_format::hex) | ||
| : std::to_chars(buf, buf + sizeof(buf), val); | ||
| ? std::to_chars(buf, buf + sizeof(buf), val, std::chars_format::hex, decimal_places) |
There was a problem hiding this comment.
You don't need to do this in the hexfloat case; full precision is assumed for hexfloats (that's sorta the whole point of them).
|
After looking at the tests, I'm not sure this is possible to do while still retaining the round-trip thing, because "0.3" is not written as "0.30000000000000004" anymore, basically. I guess this isn't going to work out after all and I'll have to maintain a fork with precision setting disabled after all. Feel free to close this, I dont' really understand how to fix this issue. Hence this change breaks all the tests, basically. Sorry for wasting your time. |
That's alright. It's kinda a damned-either-way scenario, really. Thanks for giving it a crack. |
Please do! Then I can finally remove that fork I've been using with a single line changed.. |
|
@vaartis I've implemented this in the v3 branch. val1 = 0.010284358729827818
val2 = 0.0102
val3 = 10.0102
val4 = 10.010284358729827818
val5 = 10.0using formatter = toml::toml_formatter;
std::cout << cfg << "\n\n";
std::cout << formatter{ cfg, formatter::default_flags | toml::format_flags::relaxed_float_precision };val1 = 0.010284358729827818
val2 = 0.0102
val3 = 10.0102
val4 = 10.010284358729828
val5 = 10.0
val1 = 0.0102844
val2 = 0.0102
val3 = 10.0102
val4 = 10.0103
val5 = 10.0Currently it applies to the entire formatter stream; there's no way to do it per-value. I could do that (using the same machinery I use for preserving ints as hex/dec/oct etc), though didn't bother as that seems pretty niche. Let me know if this works for you. |
marzer
added
implemented in v3
|
Thank you for implementing this. I think it's probably enough for it to be per-stream, for my use-case it certainly should be. Sort of busy with other things lately, so can't say how soon I can actually test it (since this means i'll also have to port other things like source locations to v3), but judging from your description this seems to be what I need. |
|
Tested this one too and it seems to work fine 👍 |
|
Awesome, good to hear |
2 participants
What does this change do?
Introduces a heuristic that, for floating pointer numbers that have <=3 positions after the floating point, prints them with just those, and otherwise it prints them with the usual
std::numeric_limits<T>::digits10 + 1precision.Is it related to an exisiting bug report or feature request?
Continuation of #88
Pre-merge checklist
origin/master(if necessary)