Re: [pbs-devel] applied: [PATCH proxmox-backup] api2/config/tape_backup_job: fix duplicate id parameter

[Wolfgang Bumiller <w.bumiller@proxmox.com>]

On Wed, Feb 24, 2021 at 08:28:54AM +0100, Dietmar Maurer wrote:
> Seems there is another problem with the Updater:
> 
> I can delete properties if I use normal rust naming, e.g.
> 
>    job.update_from(update, "eject_media")?;
> 
> But it does not work with kebab-case:
> 
>    job.update_from(update, "eject-media")?; // this fails silently
> 
> Please can we:
> 
>  - support kebab-case

^ that's a quick one, fix for this is in git, will bump & upload in a
bit

>  - raise error with unknown property names

^ this is a bit more involved due to flattening

One option would be to generate an enum schema for deletable fields,
though this requires a change to how we represent enums, so I can nest
the flattened objects in there.

Another would be to just generate a "string" schema with a verify
function which uses the existing property iterators to look for invalid
names. (This could probably even generate a lazy_static HashSet for
faster checks).

On Tue, Feb 23, 2021 at 06:00:33PM +0100, Thomas Lamprecht wrote:
> On 23.02.21 17:26, Dietmar Maurer wrote:
> > applied
> 
> did you read that part:
> 
> On 23.02.21 15:54, Dominik Csapak wrote:
> > i am *really* not sure if this is the correct way @Wolfgang, is
> > there another wayt to selectively use the struct members for the
> > Updater?
> 
> 
> This makes the ID optional in the schema, which is weird for an API call
> with {id} in its url (which means that without ID this can never be reached).

Now it's getting tough. Somehow I did not think about that :-)
But sure, we often need a way to identify entries, so maybe add a
`#[updater(id_property)]` attribute. (Or name it `non_optional` but that
may be too unspecific of a name.) Or are there any other reasons why a
field would behave this way?

An alternative of course would be to split the object up and flatten the
data part:

    #[api]
    struct MyType {
        id: String,
        #[serde(flatten)]
        contents: MyTypeContents,
    }

    #[api]
    struct MyTypeContents {
        <rest>
    }

But that probably gets annoying very quickly...?