ISO/TEI Conversion: Test roundtripping

[berndmoos]

• whitespace handling is not yet correct (#282, #340)
• v-tiers are duplicated -- why?

[berndmoos]

Started with a very ugly HIAT example in which utterances would end within events, overlaps were not fully specified etc. Round-tripping was very messy and effectively unusable for this example. Changed some errors/oversights in the existing XSL chain and added additional transformations, namely:

• desegment.xsl which will flatten the two-level contribution/utterance hierarchy into one level with just the contribution and utterance boundaries represented through PC
• augmentTimeline_final.xsl which will try to integrate timestamps at utterance boundaries when they are not part of the common timeline
• removeStrayAnchors.xsl which will throw out timestamps that could not be handled in the previous step

Roundtripping now seems to work for the ugly HIAT example, and it also (still) works for the more well-behaved HIAT examples: Beckhams, AnneWill and one HAMATAC transcript (Stella/Tansu). Tier order is not maintained in round-tripping because it can't be. Next: Test round-tripping for the easier cases

[berndmoos]

Roundtripping also seems to work (one test file each) for:

• FLK/FLK (one change back to the state before)
• cGATMinimal segmented (i.e. a transcription made in the PE, not in FOLKER/OrthoNormal)
• Generic segmented (some fixes unrelated to roundtripping, though)
• Event/Token

There will still be issues, probably, especially in non-timed segments and atomic timed-segments, but I can't seem to find any right now. Next step could be to make sure that all ISO/TEI exports conform to some kind of schema, and maybe also make sure that intermediate steps do so. See #369.

[berndmoos]

Let's close it until furher notice.

[berndmoos]

Further notice: Removing unnecessary timepoints is either extremely slow or it does something weird leading to no error messages, but also to the import being stuck on that step for at least 5 minutes (for a very large transcript, but so what)

[berndmoos]

It is just very slow: 24 minutes for a transcription with 8 tiers x 2250 events (FLK without tokenisation/token annotation). The imported EXB is 25 tiers x 7000 events.

[berndmoos]

Fixed an error in one of the XSLs, replaced one XSL step (remove timepoints) with Java code operating on the DOM. Also added a final "remove unused timeline items" step. It is still slow, but instead of 24 it now takes 2 minutes. Importing large, token-annotated ISO/TEI files into the Partitur-Editor is probably not the brightest of all ideas.