On Tuesday, 30 May 2017 at 22:31:50 UTC, Steven Schveighoffer wrote:
On 5/30/17 5:57 PM, Patrick Schluter wrote:
On Tuesday, 30 May 2017 at 21:18:42 UTC, Steven Schveighoffer wrote:
On 5/26/17 11:20 AM, John Colvin wrote:
On Friday, 26 May 2017 at 14:41:39 UTC, John Colvin wrote:

This version also has the advantage of being (discounting any bugs in iopipe) correct for arbitrary unicode in all common UTF encodings.

I worked a lot on making sure this works properly. However, it's
possible that there are some lingering issues.

I also did not spend much time optimizing these paths (whereas I spent a ton of time getting the utf8 line parsing as fast as it could be). Partly because finding things other than utf8 in the wild is rare, and partly because I have nothing to compare it with to know what is
possible :)

If you want UCS-2 (aka UTF-16 without surrogates) data I can give you
gigabytes of files in tmx format.

The data I can (and have) generated from UTF-8 data. I have tested my byLine parser to make sure it properly splits on "interesting" code points in all widths. UTF-16 data without surrogates should probably work fine. I haven't tuned it though like I tuned the UTF-8 version. Is there a memchr for wide characters? ;)

What I really haven't done is compared my line parsing code with multi-code-unit delimiters against one that can do the same thing. I know Phobos and C FILE * really can't do it. I haven't really looked at all in C++, so I should probably look there before giving up.


In any case, you can download the dataset from [1] if you like. There are several 100 Mb big zip files containing a collection of tmx files (translation memory exchange) with European Legislation. The files contain multi-alignment texts in up to 24 languages. The files are encoded in UCS-2 little-endian. I know for a fact (because I compiled the data) that they don't contain characters outside of the BMP. The data is public and can be used freely (as in beer). When I get some time, I will try to port the java app that is distributed with it to D (partially done yet).

[1]: https://ec.europa.eu/jrc/en/language-technologies/dgt-translation-memory

Reply via email to