Re: Maven search that uses the online APIs instead of the local index

>For the start, or if nothing else, i would suggest we change the api so
that you can access the queries and RepositoryQueries.Result and see how
far you get with other queries ...

Yes, please. I also need access to the Accessor so I can instantiate Result.

>a bit ‘heavy’ - do you see this as a principial problem or did you already
run into particular performance problems?

The online Maven API doesn't provide as much data as you could query from
Lucene. So, I would have to make multiple requests in order to get the
artifacts and then for each artifact all the versions. But this is largely
not necessary since I doubt people expand all the artifact nodes to see the
versions -- only one or two nodes will be expanded. This is why
lazy-loading the children for those nodes would be better and it would
allow me to make fewer http requests.

This would imply some changes in MavenNodeFactory.

Right now a generic query only returns the latestVersion.

>- the user should be able to make the choice if the online or offline
index should be used

Yes, some settings panel would be nice for this.

>- not all repos provide the search service

I am explicitly using this only for 'central'.

> - how to deal with cases when the network isn't available

I can probably add some icon in the status bar then.

I would also add the repository as 'skipped' in RepositoryQueries.Result.
This will be shown in the UI as containing partial results.

--emi



On Mon, Jan 11, 2016 at 3:18 PM, Tomas Stupka <[email protected]>
wrote:

> Hi,
>
> we looked into it some time ago and weren’t satisfied with how the api
> worked. Can’t remember out of the top of my head, but from a brief look
> right now it seems that all (or most) of the queries could be covered by
> the api
>
> > Let me know what do you think so far and how could we get the existing
> maven.indexer APIs to a point where I can at least publish the plugin in
> the Plugin Portal for testing.
> For the start, or if nothing else, i would suggest we change the api so
> that you can access the queries and RepositoryQueries.Result and see how
> far you get with other queries ...
> On the way we can get a better idea what other areas of the IDE it will
> impact. (as mentioned bellow)
>
> > * the Add Dependency UI and logic is a bit 'heavy' compared to the
> online APIs. For example, it loads all the versions for a given artifact at
> once instead of lazy loading the versions only when a node is expanded (see
> MavenNodeFactory). Lazy loading would map better to a 2nd search using the
> online API.
> a bit ‘heavy’ - do you see this as a principial problem or did you already
> run into particular performance problems?
> feel free to file a task for this (patches are, of course, welcome)
>
> > * existing code assumes a single GenericFindQuery service provider (I
> fixed this by adding a composite service on the 1st position)
> on the long turn we would have to figure out a better solution. Also that
> there will be more problems we would have to resolve in the maven
> infrastructure and UI if this has to work properly.
> - the user should be able to make the choice if the online or offline
> index should be used
> - not all repos provide the search service
> - how to deal with cases when the network isn't available
> - etc.
>
> regards
> t.
>
>
>