`indirect call to null` error

[axelboc]

A user of myHDF5 sent me this file: 20250829_113923_frame-altimeter.zip

When I open it in production, at myhdf5.hfgroup.org, I get an "indirect call to null" error when I select the height_min_1 dataset (all the other datasets work fine):

The error seems to come from the read_write_dataset_data function in src/hdf5_util.cc (called from get_dataset_data). My guess is, it has something to do with the chunk size being greater than the dataset size. If that's the case, it would be great to handle this more gracefully if possible.

The error only occurs in production, not in development.

Second issue in development fixed by upgrading to h5wasm@0.8.11

The fun part is that when I start myHDF5 or the h5wasm demo in development, I get a completely different error and on a different dataset!

For some reason, height_min_1 now works fine:

... but timestamp_2 no longer does, and the error thrown is different — one of the diagnostic lines reads "H5D__chunk_read(): unable to read raw data chunk":

And funnily, once I visit timestamp_2, if I go back to any other dataset, I get an "index out of bounds" error, which seems to come from the get_dataset_metadata function in src/hdf5_util.cc:

Any idea what's going on?

[bmaranville]

The second error might be related to #117, because timestamp_2 is using a scaleoffset filter with > 16 parameters for cd_values. That could lead to memory corruption.

I'm not sure about the first one - will take more investigation.

[axelboc]

Indeed, the second error is solved by upgrading to 0.8.11, sorry for the noise!

I've updated the PR description to keep the focus on the first issue. I can confirm that it occurs even with h5wasm@0.8.11, and only in production, not in development.

[bmaranville]

What is different about your production environment? I am not able to reproduce the error using the access methods in the h5wasm TS API (reading all the datasets works fine). The error I saw in production was "function signature mismatch" for the height_min_1 dataset.

I think the @h5web/h5wasm library doesn't use the h5wasm TypeScript API, correct? I would like to inspect the values being passed directly to Module.read_write_dataset_data, but I'm not sure where to look in the @h5web code at runtime.

[axelboc]

What is different about your production environment?

All I can think of is that the code gets minified and bundled by Vite.

I'm not sure where to look in the @h5web code at runtime.

The code that invokes h5wasm is in the web worker and corresponding utility file. It starts in the getValue method where it instantiates a Dataset object, then it reads the value of the dataset in readSelectedValue.

I think the error is triggered in that value getter. With the Firefox debugger, I'm able to trace the execution down to this call to Module._free().

[axelboc]

I guess React behaves differently in production as well, and of course the TypeScript and JSX code gets transpiled to vanilla JS (I don't think Vite handles this exactly the same way in development).

[bmaranville]

If you want to try h5wasm 0.8.11 with debug symbols turned on in the build, here is a copy:

h5wasm-0.8.11.zip

[axelboc]

Not sure what I did wrong last week, but it seems to be working fine after all with 0.8.11 even in production mode. 🤷 My bad, thanks for your time as always. 🙏