Dear Scott, On 2016-09-12, Scott Kostyshak wrote:
> OK let's both run the full tests after that and compare again to see if > we have any differences. I can't run the full tests (no chinese, no japanese, no hebrew, ...; no disk space, no computer to run all night). > At least for the normal (e.g. not the suspicious or unreliable) tests we > are approaching zero. Which tests still fail? The idea is, to add failures that cannot be solved immediately to "invertedTests" (currently called "suspiciousTests"): - If the cause is known, under the relevant sublabel. (Normally, there is no need for new sublabels - at least not for just one pattern.) - If the cause is not known, under the sublabel TODO (Adding the date and git-hash of the first occurence of the failure will help in tracking down the cause later.) This way, it should be easy to reach 0 failing reliable tests. (The remaining task will be to clean up the TODO section every now and than...) > Once we get there, I think it will be easy to > maintain. A big step towards "easy to maintain" will be storing of LaTeX-log output in the test log. Currently, we always have to "hand-compile" (and in many cases edit the sources before) to find out what went wrong. Günter
