Typos speller evaluation + README badge

[snomos]

For many languages (and all production languages) we have a collection of typos + expected correction. We have used this list in the past for one of several speller evalutations. It is in principle straightforward to run: send typos in, compare suggestions to expected, and calculate various statistics.

The process requires relatively little processing, is reasonably fast using divvunspell, and is thus ok to run on each push as part of a build pipeline (another pipeline usecase @bbqsrc ).

What I suggest is something along the following lines:

• generate a table with speller output along the lines of what we have done in the past (there is a new implementation in the divvunspell repo that is even faster), and store it in a document linked to from the README. It always shows the status of the last run (with a link to the commit that triggered the run). The document should be in markdown, so renderable by github on the fly.
• another document with the basic statistics for each run, so that we can follow the changes over time; each datapoint/run is identified with the commit hash. This can be a simple text file that is just appended to and automatically committed or something
• a markdown document containing an svg(?) graph generated from the data in the previous point, linked to from the README (or maybe the graph could be in the document mentioned in the first point, or in both documents at different scales)
• an svg badge embeddable in the README showing the statistics for the last run, with some colour coding for speller quality (e.g. traffic light coding for bad, acceptable and good performance).

To avoid circularity (one push triggers the test above, which creates new or updated documents, which are then committed, thus triggering a new run, etc) there should be a test that changes to the doc/ dir does not trigger a speller test run (assuming the new/updated documents are stored in the doc/ dir).

When this is in place, then do similar things for other tools in addition to spellers.

[snomos]

Until the pipelines are working, creating the statistics could be a manual process triggered by a separate make target.

[flammie]

the markdown statistics could be stored in github pages docs/, I've done this for omorfi like https://flammie.github.io/omorfi/statistics.html. This can be eventually automated neatly I hope but I have not found clean example of such github pages automation, there's a ton of actions for it but they all look overly complicated for generating few pages.

[snomos]

Nice statistics page! Ours does not have to be as detailed and rich as that one, but if it could be automatised one way or another then why not.

[snomos]

See also https://github.com/divvun/divvunspell and the section on speller testing. It presently renders output like the following:

As seen in the screen shot, the speller testing is fast, testing 1315 words in just a fraction more than a second.