Bug Tracker

I know it is documented behaviour, and I've granted as much in the report.

My point is that it would not be difficult to get an accurate count even when \glsreset is used. Provided all the data is stored at end document, and the count is not reset.

Again, I may well be missing reasons why things are how they are, I'm just starting to get better acquainted with glossaries. I just brought an idea for your consideration, if it's a good one, it's up to you. :)

Best regards,
Gustavo.

I'm sorry if I sounded a little terse. I've spent some months working on a major new version, and I'm getting quite tired and need to move onto more urgent tasks. If I get time I'll look into it before I release the pending version otherwise it will have to wait until the next version.

Oh, I'm sorry if I sounded like complaining.

I was not "picked". I was really just trying to get the point across, precisely because I know your time is scarce, and I'm aware the report was long.

So, I have absolutely no issue with however you choose to deal with the report.

Given your comment, if I may suggest, perhaps you might wish to separate the two things. Altogether, or just their "timing". I'd say the conditional writing to the .aux file is probably the thing with more relevant implications and, as far as I can see, storing all the data would be a smooth move. Besides there's no easy workaround at hand. Not resetting the count on \glsreset, on the other hand, while I do think would be a good idea, does have clear backward compatibility consequences. And, as you mentioned in the evaluation of the report, there's an easy workaround.

Again, all for your use as you see fit, when you see fit, if you see fit. I'm here in the contributing spirit, not the complaining one. :)

Thank you :-)

I thank you! :-)

Hi Nicola,

reporting on v1.48b testing, as requested.

I've seen the new \ifglsxtrresetcurrcount and corresponding switches, tried them out, and it's looking good. Thank you very much!

(I'm still hoping you consider the case of the data storage at end of document though).

Best regards,
Gustavo.

Hi Nicola,

I've been thinking again about this one, particularly my suggestion to store data of all entries at end of document. And, reconsidering, I can see cases where this might not be a good idea. If a user has a large set of entries for multiple documents, and uses a fraction of those in a given document, it would not be a good idea to store data on every defined entry. It appears to be a common practice, but I didn't think of it in suggesting. If that was what was holding your hand on the matter, you got a point.

But I think I had a little interesting idea for the problem. Considering that the variable which stores the count for a given entry is created on the fly as needed, we can know if a given entry "has been counted" in a document by testing the existence of the variable. For unit counting the following appears to work:

\documentclass[oneside]{book}

\usepackage[
  abbreviations,
  shortcuts=abbr,
]{glossaries-extra}

\makeglossaries

\GlsXtrEnableEntryUnitCounting{abbreviation}{1}{chapter}
\AddToHook{cmd/chapter/before}{\glsresetall}
\glsxtrresetcurrcountfalse

\makeatletter
\renewcommand*{\@gls@write@entryunitcounts}{%
  \immediate\write\@auxout
    {\string\providecommand*{\string\@gls@entry@unitcount}[3]{}}%
  \count@=0\relax
  \forallglsentries{\@glsentry}{%
    \glshasattribute{\@glsentry}{unitcount}%
    {%
      \ifcsundef{glo@\glsdetoklabel{\@glsentry}@unitlist}%
      {}%
      {%
        \forlistcsloop
          {\@gls@write@entryunitcounts@do}%
          {glo@\glsdetoklabel{\@glsentry}@unitlist}%
      }%
      \advance\count@ by \@ne
    }%
    {}%
  }%
  \ifnum\count@=0
    \GlossariesExtraWarningNoLine{Entry counting has been enabled
     \MessageBreak with \string\glsenableentryunitcount\space but the
     \MessageBreak attribute `unitcount' hasn't
     \MessageBreak been assigned to any of the defined
     \MessageBreak entries}%
  \fi
}
\makeatother

\newabbreviation{ABC}{ABC}{Aaaa Bbbb Cccc}
\newabbreviation{DEF}{DEF}{Dddd Eeee Ffff}
\newabbreviation{GHI}{GHI}{Gggg Hhhh Iiii}
\newabbreviation{JKL}{JKL}{Jjjj Kkkk Llll}

\begin{document}

\chapter{Chapter 1}

\ab{ABC}, \ab{ABC}, \ab{ABC}

\ab{DEF}

\ab{GHI}

\chapter{Chapter 2}

\ab{ABC}

\ab{DEF}, \ab{DEF}, \ab{DEF}

\ab{GHI}

\chapter{Chapter 3}

\ab{ABC}

\ab{DEF}

\ab{GHI}, \ab{GHI}, \ab{GHI}

\printabbreviations

\end{document}

Something similar could be done for \@gls@write@entrycounts, but using glo@\glsdetoklabel{#1}@currcount instead.

Well, just an idea, for your consideration.

Best regards,
Gustavo.

I’m very reluctant to add extra information for each defined entry in the aux file. I’ve added some extra example files in the glossaries performance page in gallery. In particular, those in Section 8 show a significant increase in overall build times for the large examples with complex builds.

Consider the makeindex 4000/6000 example that jumps from 53 seconds to 8.5 minutes simply by iterating over all entries and adding information to the aux file for every entry marked as used. (For reference, the bib2gls user manual has 4,506 glossary entries, so it’s not infeasible to consider documents with over 4000 entries.)

I think I will probably set \glsxtrresetcurrcountfalse as the default in v1.49 as you’re right it doesn’t make much sense to reset the count when the entry’s first use flag is unset.

Hi Nicola,

about \glsxtrresetcurrcountfalse, thank you.

Regarding:

I’m very reluctant to add extra information for each defined entry in the aux file.

I understand that's what you actually tested there in Section 8. And I understand your reluctance, but I'll offer respectful counterpoint, in "brainstorming" spirit, and will simply respect whatever your call is after that.

Granted, "all defined entries" (actually, only the subset which contains entrycount is being considered already) should not get their data (possibly with many empty values) written to the .aux file. But if an user enabled counting, they would arguably want the data of entries which have been used in the document to be counted, even if there is a toll to it. Currently, glossaries-extra trims the entries by the state of their first used flag at end of document. What I suggested is to write the data for entries which have had their flags set at least once along the document. It is, obviously, a larger set. But, choosing the data to store by the state of the flag at an arbitrary point (end of document) discards information which is desirable, and cripples the counting feature somewhat, because you can no longer reset entries along the document, unless you can make sure they will somehow be unset again later on (or by running \glsunsetall which defeats any saving which you are trying to achieve). Sure, there's a cost, but that's the way the feature works. And I'm telling you this being an user who does care about the matter. One of the reasons I've chosen glossaries over acro was performance. The latter would have been more than sufficient for the technical needs of my current working project, but once I had things set up my compilation time for the document had trebled, and had become a bottleneck.

In sum, as things currently are, one should "avoid resetting" with entry counting. If one does obey this documented recommendation, the set of entries which gets written to the .aux file is identical to the set of entries which has "been used at least once along the document" because, if one has not reset any of them, they are unset at end of document. So, the cost would be the same. In other words, my suggestion of storing the data of every entry which has been used at least once in the document is just a way of making entry counting compatible with the use of reset along the document. If one does not obey the recommendation, what one gets is a somewhat smaller cost (still much larger than not enabling the counting feature), and potentially unreliable counting.

That's it. My best stab at it. :-)
Again, for your use, as you see fit, if you see fit.

Best regards,
Gustavo.

Have you considered using bib2gls with record counting instead? It's more reliable because it doesn't reference the first use flag but instead relies on the number of times an entry has been indexed (recorded). For example:

\documentclass[oneside]{book}
\begin{filecontents*}{\jobname.bib}
@abbreviation{ABC,short={ABC},long={Aaaa Bbbb Cccc}}
@abbreviation{DEF,short={DEF},long={Dddd Eeee Ffff}}
@abbreviation{GHI,short={GHI},long={Gggg Hhhh Iiii}}
\end{filecontents*}

\usepackage[
  record,
  abbreviations,
  shortcuts=abbr,
]{glossaries-extra}

\GlsXtrLoadResources[loc-counters=page]

\AddToHook{cmd/chapter/before}{\glsresetall}

\renewcommand{\ab}{\rgls}
\glsxtrenablerecordcount
\GlsXtrSetRecordCountAttribute{abbreviation}{1}

\renewcommand{\glslinkpresetkeys}{\glsadd[counter=chapter]{\glslabel}}

\renewcommand*{\glsxtrrecordtriggervalue}[1]{%
 \GlsXtrLocationRecordCount{#1}{chapter}{\thechapter}%
}

\begin{document}

\chapter{Chapter 1}

\ab{ABC}, \ab{ABC}, \ab{ABC}

\ab{DEF}

\ab{GHI}

\chapter{Chapter 2}

\ab{ABC}

\ab{DEF}, \ab{DEF}, \ab{DEF}

\ab{GHI}

\chapter{Chapter 3}

\ab{ABC}

\ab{DEF}

\ab{GHI}, \ab{GHI}, \ab{GHI}

\printunsrtabbreviations

\end{document}

pdflatex myDoc
bib2gls --record-count-unit myDoc
pdflatex myDoc

Note that if you want to use hyperref, you'll need to use \theHchapter instead of \thechapter:

\renewcommand*{\glsxtrrecordtriggervalue}[1]{%
 \GlsXtrLocationRecordCount{#1}{chapter}{\theHchapter}%
}

This won't consider any instances of \glsadd unless you set the counter to chapter. For example:

\chapter{Chapter 1}

\ab{ABC}, \ab{ABC}, \ab{ABC}

\ab{DEF}

\ab{GHI}\glsadd{GHI}

\chapter{Chapter 1}

\ab{ABC}, \ab{ABC}, \ab{ABC}

\ab{DEF}

\ab{GHI} \glsxtrshort{GHI}

Hi Nicola,

Have you considered using bib2gls with record counting instead?

The things I suggested I did because the issue seemed relevant while I was trying things out, in thinking they might be a good idea for the package.

But, as I promised, I won't insist (further). ;-)
I thank you very much for the careful consideration granted to the suggestions. And I really do understand, and appreciate, the reasons for your reluctance on the matter.

Best regards,
Gustavo.

Hi Nicola,
once again, thank you!
Gustavo.