Hi Nicolas,
>>> IOW, it cannot tell the difference between a successful export and an >>> export failure with an already existing PDFFILE. >> >> This is not true as this code checks for the `errors' variable in all >> cases. With an already existing PDFFILE, you will end up with this >> message: 'Process completed with errors: ...'. > > If "file.pdf" exists before the export, you will always get "Process > completed", even if the current export was a total failure (e.g., no > file produced). > > IIUC, you're really looking after a way to know if a pdf file was really > produced. Reporting "Process completed with errors : [unknown error]" > will certainly not help on this you because some errors are not fatal > (i.e., they are skipped and the pdf file is still produced). > >> From my point of view, the issue comes from the fact that the `errors' >> variable is not correctly filled in with errors from the LaTeX log file. > > [...] > >> While the wikibooks reference >> (http://en.wikibooks.org/wiki/LaTeX/Errors_and_Warnings) tells that to >> be sure to catch *all* errors, we have to check for any line beginning >> with '!'. > > I agree, but this is not sufficient, see below. > >> Then, in the case where the `errors' variable would effectively contain >> any error from the log file, the code you mention above would work in >> any case. >> >> That's why I started with this patch (*and it works*): > > It depends on what you define as "working". We're talking about two > different things. I think a better error system should report: > > 1. a PDF file not produced (or updated), > 2. a PDF file produced with errors, > 3. a PDF file produced with warnings (maybe), > 4. a PDF file produced cleanly. > > 4 already works. Your patch improves 2, but 1 is still wrong. OK, I understand your point: the whole process has to be reviewed and improved. But, at least, my patch (or something similar) goes in the right direction providing a better catching of errors from the log file. Isn't it? Could we at least apply this patch (or a better one) that fixes this single precise issue (catching all errors from the log file)? Thanks for your help. Regards, Francesco