Tommaso Cucinotta schreef:
rgheck wrote:
I just tried, and we have segfault after 3) (infinite recursion in
Buffer::masterBuffer() [1]). Probably such operation attempt should
give up with an error to the user (but it may not be easy to
identify an arbitrary "loop" in what is supposed to be a DAG, with
documents that are not necessarily open).
There's an attempt to deal with this kind of problem in other places,
such as InsetInclude. Apparently, it pops up here, too.
Probably the parent() relationship should not be allowed to have loops
anyway. One possible way to deal with it may be to just check for
loops every time a new child->parent relationship is created, and
forbid its creation in case it creates a loop, perhaps with a warning
message to the user saying that he/she has to manually break the chain
of inclusions (and mention the document file names), otherwise LyX may
not work properly.
Yes, that will be helping a lot (actually I was already starting to do
that, see also r32378). It fixes also a problematic bug in branch
(#6251), and the recursive include you mention.
it seems the master changes according to the last-loaded master.
Also, when closing one of the two masters, its child is also closed.
This is expected, basically, at least the first part. That the child
closes whichever master closes is wrong.
Here I'm not understanding: why are children closed when their master
is closed ? Suppose I'm working on a.lyx and b.lyx, with a including
b. After a while, I decide only b needs further changes, then I save
and close a, but also b is closed -- this puzzles me a bit.
I guess the rationale is that we open all children automatically too.
So, if you're working on a multi-part document, you'd only have to open
the master to have all children, but then it's convenient that if you
close the master you'll also close all children.
So, if you can think of an intuitive way of handling this.. let us know.
Maybe, we should only close the hidden children (the ones that were not
explicitly opened by the user). Maybe, we should ask the user whether he
wants to close all children (and have a 'do not ask again' checkbox).
Maybe, we should have two options for closing a master buffer e.g.,
Close Master and Close Family.
T.
Vincent
