Each buffer now has an id which is increased when it is marked dirty
(or when one of its relatives is marked dirty).
This can be a big win since updateMacros is very expensive.
The statistics code is known to be very slow, because it relies on
DocIterator to go through the buffer.
This commit introduces a new Statistics class that encapsulates the
main code, along a virtual method Inset::updateStatistics() that
allows to fine-tune how counting is done inset by inset.
This is a faithful bug-for-bug reimplementation.
The new code appears to be 3x faster than the old one.
See bug #12929 for a discussion about statistics update woes.
The problem is caused by removing temporary aux, etc, files while
we are in the process of exporting. As Jürgen suggested, we really
do not need to remove these files until we are preparing to export:
They are not used internally. So now we just schedule the removal
and do it then.
* put the code that is called both from Buffer and InsetBrach in the
two helper methods Buffer::branchActivationStatus() and
Buffer::branchActivationDispatch().
* Cleanup the code so that _MASTER_ lfuns are disabled when there is
no master document.
* When changing branches in the master buffer, make the buffer visible
if it is not, and make sure that undo information is recorded.
* The code in Buffer::dispatch is used first, and it gives control to
the branch inset code if no branch name has been specified.
Fixes bug #12588.
Fix for bug #12456.
The labels are transmitted from Buffer to GuiRef by reference of refs_
in the getLabelList function. Previously, only one string was
transmitted. But I needed both the formatted string, e.g. "x enu:test"
or "Missing: enu:test", as well as the plain label, e.g. "enu:test".
The former is for the list of labels to choose from in GuiRef and the
latter for the label as shown in the line edit that contains the plain
label in order to create a new reference from it. Transmitting both is
what the pair achieves.
When the buffer stuff has been updated, the Buffer::needUpdate()
should be reset to false. This was forgotten in f3a0e8ff.
Fixes performance issues with MergedManuals, for example. UpdateBuffer
would be run again and again when selecting text.
Following 4a4ded22, the enabling of some change-related functions is
handled in updateBuffer. However, this method is not ran at every
document change for performance reasons.
This patch adds code to every place that modifies
Paragraph::Private::changes_ that checks whether the `changedness' of
the paragraph, err... changes.
To this end, a new helper struct is introduced that remembers
paragraph state at contruction time, and compares it to new state in
the destructor.
New forceUpdate/needUpdate methods are added to Buffer class, since
the cursor is in general not available in the places where these
changes are made.
Fixes bug #12074.
This is a reimplementation of 6d4e6aad that is both simpler and more
complete.
This uses the updateBuffer mechanism to implement a fully working
version of Inset::isChanged(). Now the function returns true for an
inset that contains an inset that contains a change, for example.
Moverover Buffer::areChangesPresent() is merely a proxy for
Buffer::inset().isChanged().
We will replace this with a better solution
For now, only keep
- Changes::isChanged()
- Buffer::areChangesPresent(), replaced by a dummy function
Next step will be to provide a working areChangesPresent() and to
compute Inset::isChanged in updateBuffer.
This reverts commit 6d4e6aad24.
When used as an adjective, both variants "descendent" and
"descendant" are acceptable, but when used as a noun only
"descendant" should be used.
For a reference, see here:
https://en.wiktionary.org/wiki/descendent#Noun
Now we report these in the same way as LaTeX errors (but let the user to
see the result anyway). It remains to be shown much is this disturbing
to users. Generally, ignoring these is not a good idea, because they are
harder to manually spot in longer documents.
The details of reported error varies because log linebreaks at 90
induced by pdflatex make log harder to parse.
The committed code is more robust than previous, in which some missing
cits/refs with long keys would go unnoticed.
Tested on bibtex and natbib.
https://www.mail-archive.com/lyx-devel@lists.lyx.org/msg208912.html
As evidenced in #11552, at loading time there is no way to tell
whether braces were added by LyX with certainty. However, after
[503f7db2/lyxgit], LyX does not automatically removes brace insets.
So, in case one faces the problems mentioned in [e8f480e7/lyxgit],
the workaround of adding a brace inset now will not be frustrated.
In essence, after [503f7db2/lyxgit], the real fix for #11552 would
have been reverting the part in [e8f480e7/lyxgit] that was adding
the braces on output.
If none of the optional arguments of a macro is used, there is no
following '[' after the macro name. Unfortunately, at loading time
the macro machinery is still not initialized, so the optionals()
member of InsetMathMacro is unusable. Hence, we have to track the
creation of macros with optionals as already we do for all macros.
The collected information is only used at loading time, because
the macros are dynamic and they may loose or acquire optional args.
Fixes the particular case reported at #11552.
This adds a new (boolean) parameter "active" to the label cache, where
we track whether a label is deleted in ct mode (the same could be done,
if wanted, for labels in notes and inactive branches).
Deleted (inactive) labels are neither considered in the uniqueness check
nor added to the outliner. This also means that undeleted references to
deleted labels are now (correctly) marked as BROKEN.
Fixes: #6563
The problem with the previous attempt was that, every time through
updateBuffer, we looked up the file location using kpsewhich, which
took too long on Windows. The new solution is to cache that info, and
to look it up only when we need it.
Previously, this info would have been re-read whenever we parsed the
bibfiles. So we re-read it now whenever the bibinfo cache is invalid,
which is less often, but should be good enough. We can add more such
re-reads if need be.
The problem with the previous attempt was that, every time through
updateBuffer, we looked up the file location using kpsewhich, which
took too long on Windows. The new solution is to cache that info, and
to look it up only when we need it.
Previously, this info would have been re-read whenever we parsed the
bibfiles. So we re-read it now whenever the bibinfo cache is invalid,
which is less often, but should be good enough. We can add more such
re-reads if need be.
The problem was that, if we killed export when some graphic was
being converted, or some external template was being handled, it
would only cancel that process, and then it would just continue.
To deal with that, we need to do a few things:
1. Modify the return values for some of the Converters routines,
and similarly the routines for external templates, so we can
tell when something has been canceled.
2. Throw an exception from InsetGraphics or InsetExternal when this
has happened during export, but ONLY when the Buffer is a clone.
We shouldn't be able to 'cancel' export when we're, say, generating
code for the preview pane, but this keeps us on the safe side..
The exception then has to be caught, obviously, back in the export
routines in Buffer.
Probably Coverity will be unhappy about something here, but I'll
deal with that problem as it arises.
Along the lines suggested by JMarc, we now collect the list of bibfiles
in use in the updateBuffer routines. This actually does simplify the code
quite a bit. See the discussion there for reasons to go this way.
Scanning is rather slow, so this improves performance in specific
situations (multiple inclusion of larger files in master/child or
chapterbib context)
