kindlegen/mobi dictionary inflections limit?

[Daltoonik]

Hey!

I'm trying to create a Kindle dictionary for a language with lots of inflections (~30 inflected forms per headword).

Everything is working fine when I am not using the full data but once I get to around 1.9M inflected forms the inflection index fails to build. (The headword index still builds fine)
I've tested the whole dataset in split up parts with the inflection index building successfully so I believe that my data is well-formed.

Info(prcgen/inflections):I1012: Compiling inflections
Info(prcgen/inflections):I3001: Number of new <idx:infl> inflection rules:  0103538
Info(prcgen/inflections):I3002: Of which rules used only once or twice:     0093970
Info(prcgen/inflections):I3003: Number of inflection rule groups:           0065714
Info(prcgen):I1019: Building index into record  0000000 Index name: 
Info(prcgen):I1025: Unicode index using optimized single-byte utf-16 encoding
Info(prcgen):I1029: The default lookup index is: ""
Info(prcgen):I1020: Building inflexions into record  0000026
Error(index build):E25002: : single entry exceeds record size (max=64k): aborting index build.
Info(prcgen):I1026: Unicode index using utf-8 encoding
Info(prcgen):I1021: Building naming index into record  0000026

(This is with the kindlegen 2.9 Linux build but it also failed with the kindlegen from the up to date Kindle Previewer)

I couldn't find any information on this error or any discussions of people having similar issues with the inflection index build. I'm wondering if there is just a hard cap on the number of inflections possible or if there might be a way to get this to build.

Thank you!

[Tiger-222]

Did you make progress since then, @Daltoonik?

[Daltoonik]

Originally Posted by Tiger-222

Did you make progress since then, @Daltoonik?

I sadly haven't had time, also a bit demotivated due to being kind of stuck.

I'm pretty sure that a non kindlegen MOBI serializer would be needed. Calibre has a custom one but I'm pretty sure that doesn't support the indicies that are used for dictionaries. Libmobi may be something that could possibly do this but since both C and working with binary formats is foreign to me I have yet to make the time investment needer for this.

There is also the concern that even if you could build the MOBI dictionary with all of those indices that maybe the kindle wouldn't be able to handle it? I mostly think that it's rather about the kindlegen software being ancient though.

For now I just have the dictionary split into 3 on my Kindle, which isn't very nice to use but is better than nothing.

[Tiger-222]

I think the error would likely be this one in case of too many inflexions (and it would not be fatal as it is only a warning):

Warning(index build):W15001: inflection rule or rule group too long (max=255). Discarded.

I have no idea about your original error yet (the same I hit too).

It is a shame that, in 2025, we have no alternative to kindlegen.