bof tap models /minutes and questions ( only some) -prototype mango annot -Jesus It looks that the port is blocked 🙁. It would be great to have this published publicly. Really interesting (in particular the very simple json profile that, I think, could be easily used by data providers). Nice talk! -Judith grouping -B Hugo: is it possible to add some more info : rotation measure , indices for radio data : FB answer: a generic mecanism for group can work with any kind of table . but you need to know the context : a priori knowledge -MT: is there a client able to interpret these groups -Judith: not yet , unfortunately . -james dempsey: how can the client know about the model annotation supported by a TAP server same for the grouping capability Discord ======= Ruby van Rooyen — Aujourd’hui à 07:43 Dear Speakers, zoom link is should be open, you are invited to join the meet Brent Miszalski — Aujourd’hui à 09:23 Is there a (science) use case for grouping TAP columns? It's not obvious to me, since the column info is already in the table TAP_SCHEMA.columns. What am I missing? LM (oral) : Compext errors, Vizier case James Tocknell — Aujourd’hui à 09:31 How would a client know if a server supports votable variants? Try and catch errors, or is that going to appear in the capabilities? Dave Morris — Aujourd’hui à 09:37 Eventually, yes it would be in the service capabilites. Jesus Salgado — Aujourd’hui à 09:33 Following Brent question, may we modify the TAP_SCHEMA to add the data model profile there? Brent Miszalski — Aujourd’hui à 09:34 Accessing the information information in the TAP tables would be preferable. Iterating over VOTable elements can be rather clunky Mark Taylor — Aujourd’hui à 09:36 Laurent, for your first suggestion of the client asking for a votable-with-mango-annotation; why not have the server just add the available annotation anyway and the client can ignore it if it is not interested? Dave Morris — Aujourd’hui à 09:38 This method leaves it open to have different data models in the future. Michel Laurent — Aujourd’hui à 09:36 Jesus, TAP allows to add table in the TAP_SCHEMA, and can add profile there ,yes Brent Miszalski — Aujourd’hui à 09:37 I also have another general TAP question. Different TAP servers use different functions for calculating distances, e.g. DISTANCE or q3c_dist (or sometimes none at all). Is there a way to work out what functions are available on the server (programmatically)? Michel Laurent — Aujourd’hui à 09:37 Brent,, yes It would be easier to do anything with tables, either table data or tap_schema tables, we just have to make sure it is doable. Jesus Salgado — Aujourd’hui à 09:37 I think now Brent is extending this proposal to prevent the annotations that is complex, I think. For the examples, to add the data model profile into the TAP_SCHEMA could be nice and clean Brent Miszalski — Aujourd’hui à 09:38 I'm not proposing anything, data model in tap schema sounds sufficient Jesus Salgado — Aujourd’hui à 09:39 OK Michel Laurent — Aujourd’hui à 09:40 Mark, yes it is the simplest way to do. The rationale for my proposal was to not confuse clients with an unexecpted mapping block, 2. The TAP SCHEMA as it is now miss some data container for properly doing the job 2. Paul Harrison — Aujourd’hui à 09:41 I think that this BOF was very useful as it did illustrate the full spectrum of possibilities from having a database that was created with no thoughts of data models and wanting to map a few columns to parts of a data model, to having a database that was created purely from a data model and wanting a faithful representation of that data model returned - I am not sure that the same solution is appropriate for both these extremes Jesus Salgado — Aujourd’hui à 09:42 Yes,... to mix the info of the data model profile into the TAP_SCHEMA would be a good way to close the loop 2. Thanks for the presentations. Really interesting BoF Dave Morris — Aujourd’hui à 09:43 Ideally, the client should get the same response structure from both cases ? Michel Laurent — Aujourd’hui à 09:43 Paul, in theory the same solution could work for the whole spectrum, but the has to be demonstrate .... WIP James Tocknell — Aujourd’hui à 09:44 Continuing on the models, how would a client know if the models haven't been applied, what the error was: e.g. the tables/columns could lack model information or the post-query filter could have failed, and the user may want to do different things depending on the error (the former there's not much the user can do, but maybe in the latter case changing the query might help)? Mark Taylor — Aujourd’hui à 09:44 OK. From my (client) side, it shouldn't be hard to ignore that information if it's present. I'd say it's reasonable to expect clients to do that, as long as VODML is compatible with (either present or future) VOTable standard. Dave Morris — Aujourd’hui à 09:44 From the client perspective the two cases should just be TAP services that provide model annotations. 2. Paul Harrison — Aujourd’hui à 09:44 2. I am not sure - pretty much the whole point of SQL as a query language is to denormalize 2. 2. 2. 2. 2. [09:45] 2. so if you do want to return model instances you really do want a OOQL Michel Laurent — Aujourd’hui à 09:45 James, there are the usual VO meta-data for this (UCD etc) and some specific skill on client side to run approrioate inferences 2. Mark, If you are right, I agree to postpone the FORMAT proposal until we have a case where different models can be applied to the same table. Mark Taylor — Aujourd’hui à 09:50 That sounds reasonable to me (though of course others might have different opinions...). In terms of using this metadata I can imagine it would be easier for a client like topcat to work with this information if it happens to be present in a query response than to have to work out what model(s) it should ask for up front. Dave Morris — Aujourd’hui à 09:50 Yes, in the long run we might indeed create use an OOQL. Is this is a useful step towards that, adding layers of abstraction. Michel Laurent — Aujourd’hui à 09:50 Paul, if you take the case of a model with source->detection->spectra and you make a selection on sources, the annotater should able to construct a responses with 3 joined table. So that you fix the denormalization issue. This requires however the query to be limited to souces (SELCT SOURCES.* FRM sources). If is not the case, the annoter left the maping empty 2. Just try this https://saada.unistra.fr/tapcomplex/ (should work now) It does a renomalization on client side Paul Harrison — Aujourd’hui à 09:55 I agree in practical terms that it is mainly the select clause that has to be interpreted differently, but there are other parts of the query that you might want to limit - e.g. how deep into the object tree is retuned Michel Laurent — Aujourd’hui à 09:57 Yes, the real problem are the loops in the model. They are broken in my client. Michel Laurent — Aujourd’hui à 10:02 The ongoing models are rather flat (MANGO/CUBE) . Provenance comes with a safe TAP mapping.But if we want to support any model, we will have to dive on some sorte of OOQL.Note I've been hired 20 years ago as OQL expert For this complex cases I would rather lean toward some parameter based langage (CGI-like) than reopening the OOQL box. Michel Laurent — Aujourd’hui à 10:05 Anyway, we can keep in touch if you want to contribute or even just follow the topic Mireille Louys — Aujourd’hui à 10:38 Hi Laurent , will you collect our bof presentations somewhere to share them for a longer time ? Michel Laurent — Aujourd’hui à 10:45 I'll open Wiki page on the IVOA