Guenter Milde wrote: > However, whenever a reference changes (to adapt to a new file version or > other changes) or a new test file + reference is added, one must also > check if the new reference is really a *good* one.
Yes, this is the key point. > In the case of LyX, this includes checking whether the representation in > the LyX GUI is OK and a test compilation. Only if "compilability" is not > achievable without disproportionally high effort, it may be sacrificed -- > but not without documenting this special case and the reasons in the test > file and reference. Here I disagree. Assuming that the .tex source is compilable, it is not enough to test whether the converted LyX file compiles as well, you also need to test the contents. Of course this can be done by visual comparison of two PDFs, but this is not the only way. It is also quite effective to compare the two .tex files (original .tex and roundtrip .lyx.tex) with a good diff tool (e.g. xxdiff or winmerge). Compilability depends on much more than tex2lyx, therefore I don't think we should require it. If we can get it almost for free for a given test case it is fine, but if not then it is no big deal IMHO. Look at the existing LyX export tests: Many of them fail for some configurations, and only LyX is involved here, no tex2lyx, so it should be much more easy to have them compilable, but in reality many of them are not. Georg
