Question about FORMAT specifier's purpose and usefulness in a DataLink-dominated world

[gpdf]

Colloquially, the FORMAT parameter is intended to apply a constraint to the persistence format of the data product associated with a row in a conceptual underlying ObsCore table. (Obviously there doesn't have to be such a table in order to implement SIAv2/DAP, but the standard is written with reference to a table model.)

From a user perspective, this is clearly meant to enable, for instance, limiting a query to data in FITS format. In the DAP era where tabular datasets may be available from a service, a user might say "I'm only interested in Parquet".

The way the standard is written, though, it's clear that FORMAT is meant to be evaluated against the value of access_format in the query response. In many archives (CADC, Rubin, probably at least parts of IRSA in the future) we have adopted the "DataLink model" for providing ObsTAP/SIAv2 dataset access, though, where the actual access_format value is always the DataLink MIME type.

The standard even acknowledges this:

This column describes the format of the response from the access_url (see 3.1.3) so the values could be data file types (e.g. application/fits) or they could be the DataLink MIME type.

It seems like we've accidentally shot ourselves in the foot here. No non-IVOA-aware science user would be expected to be know about the "DataLink model" -- they go to Firefly, say, do a query, and, if possible, Firefly will show them the #this target from the DataLink links service, rather than making them navigate the indirection on their own. Unless they deliberately click on the part of the UI that lets them see the links response and any associated additional datasets, they won't be aware of DataLink at all. That's a good thing.

So in this situation if someone does FORMAT=fits they are likely to be very surprised by the results.

I realize the difficulties involved in potentially prying FORMAT off its mandatory link to access_format, but I think it would be worth our having a conversation about whether an interpretation "if access_format is the DataLink MIME type, then evaluate the restriction against the content_type for the #this entry in the resulting links table" would be sustainable, at least as an option.

I recognize that this might require data publishers to add a column to the underlying table to make sure that the "real" data type is efficiently queryable.

FORMAT seems sufficiently useful that it's a shame to, in effect, be forced to lose its usefulness in exchange for all the other big advantages of the "DataLink model". For Rubin (IMO) it's still a worthwhile tradeoff if we can't fix this, but... let's try to think this through and fix it.

My guess is that this must have been discussed before, but I haven't found the trail yet.

[msdemlei]

On Tue, May 21, 2024 at 10:23:21AM -0700, gpdf wrote: `FORMAT` seems sufficiently useful that it's a shame to, in effect,

Hm. Is it? Useful, I mean? With all my time in the VO I've yet to find a credibly useful application. In particular I cannot believe that someone would want to have a "good" result swallowed just because it may be in a format they cannot immediately read. Also, in the rare cases in which I have put multiple formats for the same dataset into a table (in SSAP), I've always regretted it. If that is a typical experience, then FORMAT in effect becomes a parameter "either hit the sort of format the resource has and get the same result that you'd get if you left out FORMAT, or miss it and you'll have an empty result". That, again, does not seem to be useful behaviour to me. So... now that we have datalink, I'd all be for dropping FORMAT and telling people who want to disseminate their data in multiple formats to use datalink.

[pdowler]

I would support dropping it for same reasons expressed above: not actually useful and leads to surprising results.

[gpdf]

OK, I'm actually pretty glad to hear you both say this.

The recommendation to interactive clients' developers, then, might be "if #this, or any other entry in a DataLink table appears in multiple rows differing only by content-type, the client may wish to display to the user that they have a choice of available data formats for retrieving the data".

(Noting that this is not about "science FITS vs. preview JPEG" but more about "science FITS vs. science HDF5" or, say, "VOTable-TABLEDATA vs. Parquet" -- choices of content-equivalent representations.)

[gpdf]

How do we get this rolling? Is this a deprecation warning to be added to the next SIA 2.x, along with an explanatory sentence or three?

Or do we view FORMAT as permanently part of SIAv2, and we just start DAP without it from the very beginning?

[gmantele]

I have never used or implemented SIA, so my opinion may not be the best here. But I would tend to say, let's start DAP without the FORMAT constraint. Services are still free to implement it if they want to, but it won't be standard. If at some point it seems to be important to have it (which does not seem to be the case from what all of you say), then, you'll be able to do add it into DAP. In standards, it is often easier to add than to delete.

[pdowler]

Yes, I think we would just drop FORMAT from DAP entirely. Maybe we don't need to do anything with SIAv2... I can't really think of an erratum that would be at all helpful.

[gpdf]

That sounds reasonable to me.

Regarding SIAv2: The message one would like to convey is "On some archives, FORMAT is unlikely to do what you think it should do" but it almost seems like that's something that should be in client documentation rather than in the standard. If we were going to issue a new SIAv2.* for some reason, I can think of some better wording, but I don't think issuing an erratum (which we don't fold back into the source document anyway) is going to get the message out effectively.