Guile AWS is pre-alpha software.  At the very least it’s yet another demonstration that Guile’s compiler tower can be used to generate an embedded domain specific language from JSON specifications.

The DSL Guile AWS produces is unpolished and thus pretty repetitive and ugly.  Here is an example session to create an EFS and make it ready for mounting on an EC2 instance:

#+begin_src scheme
(import (aws api elasticfilesystem-2015-02-01))

(CreateFileSystem
 (CreateFileSystemRequest
  #:CreationToken (CreationToken "my-guile-aws-filesystem")))

#; (("ThroughputMode" . "bursting")
 ("Tags" . #())
 ("SizeInBytes"
  ("ValueInStandard" . 0)
  ("ValueInIA" . 0)
  ("Value" . 0)
  ("Timestamp" . null))
 ("ReplicationPolicies" . null)
 ("ProvisionedThroughputInMibps" . null)
 ("PerformanceMode" . "generalPurpose")
 ("OwnerId" . "439516136713")
 ("NumberOfMountTargets" . 0)
 ("Name" . null)
 ("LifeCycleState" . "creating")
 ("KmsKeyId" . null)
 ("FileSystemId" . "fs-8bee03d0")
 ("FileSystemArn"
  .
  "arn:aws:elasticfilesystem:eu-central-1:439516136713:file-system/fs-3c759b67")
 ("Encrypted" . #f)
 ("CreationToken" . "my-guile-aws-filesystem")
 ("CreationTime" . 1614808760.0)
 ("AvailabilityZoneName" . null)
 ("AvailabilityZoneId" . null))

(CreateAccessPoint
 (CreateAccessPointRequest
  #:ClientToken (ClientToken "my-guile-aws-filesystem")
  #:FileSystemId (FileSystemId "fs-8bee03d0")))

#;
(("Tags" . #())
 ("RootDirectory"
  ("Path" . "/")
  ("CreationInfo" . null))
 ("PosixUser" . null)
 ("OwnerId" . "439516136713")
 ("Name" . null)
 ("LifeCycleState" . "creating")
 ("FileSystemId" . "fs-8bee03d0")
 ("ClientToken" . "my-guile-aws-filesystem")
 ("AccessPointId" . "fsap-0d9a986284d086526")
 ("AccessPointArn"
  .
  "arn:aws:elasticfilesystem:eu-central-1:439516136713:access-point/fsap-0d9a986284d086526"))

;; Use the same subnet identifier as your EC2 instances.
(CreateMountTarget
 (CreateMountTargetRequest
  #:FileSystemId (FileSystemId "fs-8bee03d0")
  #:SubnetId (SubnetId "subnet-7f6a7102")))

#;
(("VpcId" . "vpc-8e31f4e4")
 ("SubnetId" . "subnet-7f6a7102")
 ("OwnerId" . "439516136713")
 ("NetworkInterfaceId" . "eni-08df70c51f2ecbc33")
 ("MountTargetId" . "fsmt-023b3e5b")
 ("LifeCycleState" . "creating")
 ("IpAddress" . "172.31.44.41")
 ("FileSystemId" . "fs-3c759b67")
 ("AvailabilityZoneName" . "eu-central-1b")
 ("AvailabilityZoneId" . "euc1-az3"))

;; Tear down
(DeleteMountTarget
 (DeleteMountTargetRequest
  #:MountTargetId (MountTargetId "fsmt-284b4e71")))

#; #t

(DeleteAccessPoint
 (DeleteAccessPointRequest
  #:AccessPointId (AccessPointId "fsap-0d9a986284d086526")))

#; #t

(DeleteFileSystem
 (DeleteFileSystemRequest
  #:FileSystemId (FileSystemId "fs-8bee03d0")))

#; #t
#+end_src

The output is pretty bad as it is currently unprocessed SXML or JSON.  It may not even work at all, because the AWS APIs are all a little different.

Considering all these caveats there are a couple of obvious things to work on:

** Extend requestUri formatting
   The =(aws request)= module interprets the =requestUri= field and substitutes placeholders. However, this was only tested with the EFS API.  The S3 API uses fancier placeholders such as ="/{Bucket}/{Key+}?restore"= — it is not clear yet how to interpret the plus.
** Create aliases
  The S3 API (for example) defines aliases for some operations, such as “PostObjectRestore” for “RestoreObject”.  The compiler should process the “alias” field.
** Record possible errors
The S3 API and the EFS API (for example) define possible error names.  While their shape is not specified anywhere we should generate values for these error conditions.
** Process output shapes
   We generate types for all defined shapes — including output shapes — but we don’t mashall the output SXML into appropriate Scheme values yet.
** Turn errors into Scheme conditions
This is easier said than done because different APIs return different kinds of errors.