This commit adds a new interface that is capable of runtime
use. By calling Request(action).call(request, options), the request
object can now be built at runtime. This change also moves the client
object into the options structure. It also moves the metaInfo generated
function to a type-based function rather than requiring an instance for
binding.
This provides initial implementation with some basic calls
working. Only GET requests have been verified so far.
Http header support missing. POST/PUT might work, but
have not been tested. Currently the demo tests are
failing due to a response deserialization issue that
is still being debugged
This will use the actual structure name or the override
from the trait as needed. Return value support
is also enabled in the code generation but not in
aws.zig. The current fuzzy checks should get most
of the way there though
Instead we will rely here on the implicit struct provided by the file
itself. This means that when the service_manifest.zig import is
assigned, we don't end up with a construct like "manifest.sts.sts" to
get to the sts service. This simplifies greatly the way that
servicemodel.zig needs to behave.
One down side, however, is that the structure does not seem
accessible with the current zig language, making metaInfo unable
to access itself as it did before. Or maybe it can, I just can't find
it. So, this change also adds a new "service_metadata" public constant
with the same declarations being published at the file level, and
that is the new return from the metaInfo function. Our aws.zig
only really needs the action and that metadata, so we're ok with that
even if there is some duplication (we could codegen pointers over, and
maybe should to save a little bit of space).
There are two recent changes in zig that effect awshttp.
1. cf65ab8 disallows unused variables. Fair enough and backward
compatible.
2. https://github.com/ziglang/zig/issues/2115#issuecomment-827968279
This comment resulted in a backward incompatible change to use the
underlying value from a C enum rather than its symbol. This reduces
edge cases in the compiler. Ultimately we may want awshttp to define
zig enums that mirror the C enums, but for now I've commented the
definitions of the C enums used.
Key changes:
1. Remove all unused variables (see zig commit cf65ab8)
2. Remove tests requiring other files
3. Specify enum type when providing specific enum values (see zig
commits around 507a8096d)
code in the function required found variable to be a real count
for allocation purposes, but it was originally coded as a flag
instead, which broke any field with more than two underscores
There appears to be a leak in json.zig, which had minimal modifications for
snake/camel case, including use of an allocator. It is not obvious that
the leaks were added by these changes, so I will leave this issue for
later investigation.
This should complete the query protocol calls. However, there are likely
gaps in implementation for the transformation of request parameters
to url encoded body data
This seems silly, but I guess in AWS it is a thing that these two values may
be different. Consumers of awshttp have the option of ignoring this, which
should be correct most of the time. Aws.zig will, however, use the service
metadata to do the right thing
This increases compilation time significantly as all 260+ services
need to be analyzed by zig during compilation. I plan to change
the model from a single "services" constant to a function
that will import only services that plan to be used. This
might be in addition to the single constant to allow
consumers to choose short compile times or all services
support
