Object schemas can now declare a field which causes
'additional_properties' to be set to true and the field being ignored
in the schema.
This allows adding a flattened HashMap<String, Value> to gather the
additional unspecified properties.
#[api(additional_properties: "rest")]
struct Something {
#[serde(flatten)]
rest: HashMap<String, Value>,
}
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
This does not "stream", but rather skips the intermediate step to
serialize the entire output into a local json string.
We now reserve the "Stream*" prefix for actual *streaming*, that is,
producing an API response which gets streamed continuously as it is
asynchronously produced.
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
separated out of FieldAttrib without the `flatten` attribute, since we
don't support this on enum variants
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
This mostly affected attribute parsing (due to the syn::Meta changes).
Also creating `DelimSpan`s for custom-built `syn::Attribute`s is a
bit... ugly.
Upshot: turns out we can drop some helpers in util.rs with the new
`syn::Meta` changes.
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
This allows fixing up things such as `skip_serialize_if`
calls like so:
#[derive(Updater)]
struct Foo {
#[serde(skip_serializing_if = "MyType::is_special")]
#[updater(serde(skip_serializing_if = "Option::is_none"))]
field: MyType,
}
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
to generate the `Streaming` variants of the ApiHandler
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
* Updatable is now named UpdaterType
* UPDATER_IS_OPTION is now assumed to always be true
While an updater can be a non-optional struct, being an
updater, all its fields are also Updaters, so after
expanding all levels of nesting, the resulting list of
fields can only contain optional values.
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
This way we can assign `API_SCHEMA` constants to `Option`
types.
Here's why:
The api-macro generated code usese `T::API_SCHEMA` when
building ObjectSchemas.
For Updaters we replace `T` with
`<T as Updatable>::Updater`
This means for "simple" wrappers like our `Authid` or
`Userid`, the ObjectSchema will try to refer to
`<Authid as Updatable>::Updater::API_SCHEMA`
which resolves to:
`Option<Authid>::API_SCHEMA`
which does not exist, for which we cannot add a normal
`impl` block to add the schema variable, since `Option` is
not "ours".
But we now have a blanket implementation of `ApiType` for
`Option<T> where T: ApiType` which just points to the
original `T::API_SCHEMA`.
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
They don't appear in the json data structure and therefore
should not be named separately in the schema. Structs with
flattened fields will become an `AllOf` schema instead.
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
to be non fatal for better error messages, this way the user
will see the compile error, but we still generate all the
code & schema variables so that one error isn't accompanied
by all the ones resulting from not having the generated code
there at all.
Eg.
error: description not allowed on external type
--> src/api2/access/user.rs:472:22
|
472 | description: "Get API token metadata (with config digest).",
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Was previously also accompanied by
error[E0425]: cannot find value `API_METHOD_READ_TOKEN` in this scope
--> src/api2/access/user.rs:774:11
|
699 | pub fn delete_token(
| ------ similarly named constant `API_METHOD_DELETE_TOKEN` defined here
...
774 | .get(&API_METHOD_READ_TOKEN)
| ^^^^^^^^^^^^^^^^^^^^^ help: a constant with a similar name exists: `API_METHOD_DELETE_TOKEN`
The second error was "wrong" and came much later, needlessly
filling the screen if this happened on multiple functions.
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
References to external schemas (or types) already include
the description in the external schema and therefore are
illegal.
The implementation consists of multiple parts:
* Introduce a `Maybe` type which can be `Explicit`,
`Derived` or `None`.
* Forbid `Explicit` descriptions on references.
* Instead of bailing out on such errors which causes all of
the generated code to vanish and create heaps of
additional nonsensical errors, add a way to *add* errors
without bailing out immediately via the `error!()` macro.
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
Support raw parameter name identifiers (eg. `r#type`)
#[api(
input: {
properties: {
type: {
type: String,
description: "Foo",
},
},
},
)]
fn foo(r#type: String) { code... }
The "r#type" parameter in the fn decl will match the "type"
parameter name in the input property list.
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
Instead of setting a default value to a const and inside an
.unwrap_or_else closure, lets set it only to the const and reuse that
later in .unwrap_or
To achieve that we move the "unrwap_or" code for param plumbing code generation
a bit later so that we have easy access to the generated const name.
As all this code is related to optional/default-value stuff it does read still
relatively OK with that change, IMO.
This has the advantage of not getting a warning like:
> warning: constant is never used: `API_METHOD_EXAMPLE_FOO_PARAM_DEFAULT_FORCE`
> --> src/api2/node/foo.rs
> |
> XY | force: {
> | ^^^^^
> = note: `#[warn(dead_code)]` on by default
When one has a API endpoint like:
> #[api(
> input: {
> properties: {
> force: {
> type: bool,
> optional: true,
> default: false,
> },
> },
> },
> ...
> )]
> /// Example
> fn example_foo(force: bool) -> Result<(), Error> {
> if force {
> // do something
> }
> Ok(())
> }
It effectively changes the output for optional parameters with a default set
and no Option<T> from
> let p = p.unwrap_or_else(|| #default_value);
to
> let p = p.unwrap_or(#const_name_for_default);
where the "#const_name_for_default" is a pub const with value
"#default_value"
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
after benchmarking (again), i found that doing a simple find instead
of saving the inidices for the ident strings in a hashmap has
no real performance impact (the max list size for the properties
are max ~25 at the moment, so this should not be impacting compile
times much) but it is much simpler
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
the 'properties_' list is sorted by the the literal string of a
fieldname, but we binary-search for the 'ident_str' (which may be
different, since we map '-' to '_' for example)
by creating a hashmap to map from ident to index, we can do a simple
lookup in that case that will work
benchmarks showed no measurable performance difference
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
