On 2015-11-10, Kornel Benko wrote: > Am Dienstag, 10. November 2015 um 10:50:30, schrieb Guenter Milde > <mi...@users.sf.net> >> On 2015-11-10, Kornel Benko wrote: >> > Am Dienstag, 10. November 2015 um 07:34:42, schrieb Guenter Milde >> > <mi...@users.sf.net>
>> for a safe handling of XeTeX + TeX-fonts without hacks in the LyX code, I >> recommend to allow this combination only, if "inputenc" == "ascii". > I did it. That way I got 117 less failed exports. > (Previously 260, now 143). Good. >> for a safe handling of LuaTeX + TeX-fonts without hacks in the LyX code, I >> recommend to allow this combination only, if "inputenc" != "auto". > Setting 'ascii' for '(auto|default)' leads to 53 failed pdf5_texF tests. > Without this change I have only 28. This does not feel right. Right. As I said below, "ascii" is not the best choice. Not well tested and some documents fail with it, (independent of the engine). >> For LuaTeX + TeX-fonts, only "auto" needs to be changed. Preferably to >> the document languages default encoding, but any 8-bit encoding or >> ascii will do. > This I am omitting for now. You could try with "latin9" instead of "ascii". Or, make it dependent on the document language - for manuals etc. this does not require looking into the LyX-file, as non-English documents are stored in directories with the language tag. A language tag to encoding mapping can be extracted from lib/languages. >> Mind, that changing the inputencoding is only seldom tested and can >> exhibit a number of currently hidden problems. For example, the Russian >> documents fail with inputenc==ascii due to #9637 (textgreek and textcyr >> depend on font-encoding, not input encoding) where the spurious \textcyr >> commands interfere with ERT in the document. >> OTOH, the utf8 inputencoding fails with Greek and Russian due to >> #9681 (textgreek and textcyr also required for encodable characters). >> I therefore recommend also test exporting documents with pdflatex and >> inputenc=ascii as well as inputenc=utf8. > Now it starts to be complex. It means to analyse the lyx file before test. > We are not doing it yet. What I have in mind here was a number of additional tests, where all manuals (say) get "inputencoding" set to "ascii" and tested for export to pdf2, say. +1 if a XeTeX-TeXF test fails, we can find out if this is due to XeTeX vs. 8-bit LaTeX or to inputenc set to "ascii". -1 we get a number of new failing tests. Similar, I would add test for manuals with "inputencoding" set to "utf8" and export to pdf2. Günter