docx creates nested runs (<w:r><w:t><w:r><w:t>), which are then invisible in the opened document

[unhammer]

$ for y in yes no; do APERTIUM_TRANSFUSE=$y apertium -f docx -u -d . nob-nno  /tmp/in.docx >/tmp/ut.$y.docx; done

in.docx

With transfuse, we get this bit:

      <w:r>
        <w:rPr/>
        <w:t xml:space="preserve">
          <w:r>
            <w:rPr/>
            <w:t xml:space="preserve">Dette er såkalla «Sideloaded Add-ins». Dei nyttar eit webview, i praksis ein nettlesar</w:t>
          </w:r>
        </w:t>
      </w:r>

which word (and libreoffice) don't show on opening the document, presumably nested runs aren't allowed in OOXML.

(Note: If I first save in.docx from Libreoffice, transfuse can handle it fine, because LO merges all the runs in the input paragraph on saving (removing the proofErr stuff).)

[unhammer]

Strangely, the text isn't divided in input to the pipeline. The split-point is nettleser som, and input xml has

        <w:t xml:space="preserve">». De benytter et webview, i praksis en nettleser som er bygget inn i Office-programmene, for å vise innholdet sitt og utføre oppgavene sine. </w:t>

Here's what it looks like for the first step of the pipeline:

[transfuse:\/tmp\/transfuse-D7x3hD-8b_Y]

[tf-block:1-Zh6TUA]

Teknologi.[]

[tf-block:2-SKmwAw]

[[t:text:SyTAKg]]Dette er såkalte «Sideloaded Add-ins». De benytter et webview, i praksis en nettleser som er bygget inn i Office-programmene, for å vise innholdet sitt og utføre oppgavene sine.[[/]] .[]

Then right after the full pipeline, we have wordbound tags galore:

[transfuse:\/tmp\/transfuse-D7x3hD-8b_Y]

[tf-block:1-Zh6TUA]

Teknologi.[]

[tf-block:2-SKmwAw]

[[t:text:SyTAKg]]Dette[[/]] [[t:text:SyTAKg]]er[[/]] [[t:text:SyTAKg]]såkalla[[/]] [[t:text:SyTAKg]]«[[/]][[t:text:SyTAKg]]Sideloaded[[/]] [[t:text:SyTAKg]]Add-[[/]][[t:text:SyTAKg]]ins[[/]][[t:text:SyTAKg]]»[[/]][[t:text:SyTAKg]].[[/]] [[t:text:SyTAKg]]Dei[[/]] [[t:text:SyTAKg]]nyttar[[/]] [[t:text:SyTAKg]]eit[[/]] [[t:text:SyTAKg]]webview[[/]][[t:text:SyTAKg]],[[/]] [[t:text:SyTAKg]]i[[/]] [[t:text:SyTAKg]]praksis[[/]] [[t:text:SyTAKg]]ein[[/]] [[t:text:SyTAKg]]ne[[t:text:SyTAKg]]ttl[[/]]esar[[/]] [[t:text:SyTAKg]]som[[/]] [[t:text:SyTAKg]]er[[/]] [[t:text:SyTAKg]]bygd inn[[/]] [[t:text:SyTAKg]]i[[/]] [[t:text:SyTAKg]]Office-[[/]][[t:text:SyTAKg]]programma[[/]][[t:text:SyTAKg]],[[/]] [[t:text:SyTAKg]]for[[/]] [[t:text:SyTAKg]]å[[/]] [[t:text:SyTAKg]]visa[[/]] [[t:text:SyTAKg]]innhaldet[[/]] [[t:text:SyTAKg]]sitt[[/]] [[t:text:SyTAKg]]og[[/]] [[t:text:SyTAKg]]utføra[[/]] [[t:text:SyTAKg]]oppgåvene[[/]] [[t:text:SyTAKg]]sine[[/]][[t:text:SyTAKg]].[[/]] .[]

The second-to-last step, before postgenerator, looks like
[[t:text:SyTAKg]]ne~tt[[/]][[t:text:SyTAKg]]lesar[[/]]
at the split-point, while after postgenerator we get
[[t:text:SyTAKg]]ne[[t:text:SyTAKg]]ttl[[/]]esar[[/]] [[t:text:SyTAKg]]som[[/]]

So is the issue here that postgenerator should not be creating these nested word blanks, or that transfuse should somehow know how to deal with nested word blanks?

[unhammer]

@mr-martian does your apertium/lttoolbox#144 avoid nested word blanks in postgen?

[TinoDidriksen]

The pipe may not yield nested structures, nor will Transfuse give it nested structures, so that looks like a bug in postgen.