Commits 8f48f91009 (ODT export: omit
unreferenced <text:list xml:id="...">, 2022-03-10) and
82bbf63582 (tdf#155823: Improve the
check if the list id is not required, 2023-06-14) tried to improve
deterministic ODF output, by omitting the list identifiers in case
when those identifiers were unreferenced. The latter of these used
document model node numbers to check if other lists appeared after
the last occurrence of the list that is continuing in the current
node. But it turned out, that this isn't robust. Consider this ODF:
<text:list xml:id="list1" text:style-name="L1">
<text:list-item>
<text:p>a</text:p>
</text:list-item>
</text:list>
<text:p>b<text:note text:id="ftn1" text:note-class="endnote"><text:note-citation>i</text:note-citation><text:note-body>
<text:list text:style-name="L2">
<text:list-item>
<text:p>x</text:p>
</text:list-item>
</text:list></text:note-body></text:note></text:p>
<text:list text:continue-list="list1" text:style-name="L1">
<text:list-item>
<text:p>c</text:p>
</text:list-item>
</text:list>
The paragraphs a, b, and c are all in the main document body, and
have sequential document model node numbers (say, 15, 16, 17). If
these numbers are checked, there is no node between node 15 ("a")
and node 17 ("c") with a different list (both 15 and 17 belong to
a list with style "L1" and identifier "list1", and node 16 doesn't
belong to any lists). That suggests that the list identifier isn't
needed in this case. Bug when the actual output of node 16 is done,
it includes a node from an endnote ("x"), which is located in a
different place in the document model, and has a node number like
7 (so not between 15 and 17). The paragraph "x" belongs to another
list with style "L2", and is output to ODF between paragraphs "a"
and "c". Here, we must refer from paragraph "c" to the list of the
paragraph "a" using the list id, but this is not obvious when only
considering node numbers, and requires the prior knowledge of the
actual order of appearance of lists in the ODF.
Unless we build a DOM, this is only possible, if we do a two-pass
output, and collect the nodes order in the first pass. The output
already does that in a "collect autostyles" pass. The problem here
is that the "collect autostyles" pass used an optimized function,
XMLTextParagraphExport::collectTextAutoStylesOptimized, introduced
in commit 8195d7061e (INTEGRATION:
CWS swautomatic01 (1.126.4); FILE MERGED, 2006-12-01) for #i65476#
and which used style::XAutoStylesSupplier for optimization to get
the autostyles.
This drops XMLTextParagraphExport::collectTextAutoStylesOptimized,
and reverts to use of collectTextAutoStyles, which handles nodes
in the same order as when writing to ODF. There, we build a vector
of the node numbers sequence, used later to sort DocumentListNodes.
This uncovered an omission from the work on paragraph mark (commit
1a88efa8e0 tdf#155238: Reimplement
how ListAutoFormat is stored to ODF, 2023-05-11). Turns out, that
the code in SwTextFormatter::NewNumberPortion introduced in commit
cb0e1b52d6 (sw, numbering portion
format: consider full-para char formats as well, 2022-10-20) was
left behind when re-implementing paragraph marks to use dedicated
property; empty trailing spans still affected how the lists were
rendered, and that allowed to overlook import defects, where the
paragraph mark properties weren't properly set.
In ODF import (XMLParaContext::endFastElement), for compatibility,
this treats empty trailing spans as defining paragraph mark (when
the paragraph mark wasn't set explicitly). This way, the trailing
spans get converted to the paragraph mark.
In WW8 import, last cell paragraphs didn't call the code handling
the paragraph marks. This is also fixed now.
The changes result in slightly different numbering of autostyles
in the ODF. It seems, that the new numbering more closely follows
the order of appearance of the autostyles in the output; and some
cases of autostyles that were written, but unreferenced, are now
eliminated. The unit tests were updated accordingly.
I hope that the performance impact on the export time would not be
too large.
It is unclear why outline numbering exports a list element at all.
Fixing that to not emit the list element is a separate task / TODO.
Change-Id: I5c99f8d48be77c4454ffac6ffa9f5babfe0d4909
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/166572
Tested-by: Jenkins
Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com>
69ed893087