rgheck wrote:
Pavel Sanda wrote:
puzzled I got about why such extensive changes had to be made and
couldn't be confined to some relatively well defined aspect of the
code.
...
the whole implementation was nearly impossible to understand.
it is still the case or its better now?
It's still as much of a mess as it was.
do i also understand correctly that the changes pervade to larger
areas of
code which is rather far away from bundling? (sorry for such novice
questions,
i havent followed the threads closely, nor i read the code)
Yes. Every inset that references a file that might be bundled has
undergone extensive changes as a result. This is what led to the
present situation, where existing LyX files that reference, say,
BibTeX databases via a simple name like "biblio" don't work. And the
really disturbing thing is that they don't work when you're not using
embedding. If you are using embedding, they work fine. This is because
the EmbeddedFile architecture is deeply entwined with the way insets
operate, so much so that changes in how EmbeddedFile is implemented
can lead (and have led) to bugs in the inset code, even in cases where
embedding is not being used.
To give an example: So far as I can see, the EmbeddedFile stuff works
almost entirely via the GUI. So changes that got made via the
mini-buffer don't interact with the EmbeddedFile stuff, and this can
lead to inconsistencies that could, in principle, cause all kinds of
problems, including crashes. Despite extensive study of the code, I'm
quite unable to be sure whether there are such problems or not. But the
fact that the file copying that needs to happen to make the embedding
stuff work happens in places like GuiGraphics makes me nervous.
Richard
