fix: match batch response parts to requests by Content-ID - #2726
Conversation
Batch::parseResponse() paired each response part with a request by position ($requests[$i-1]), and silently ignored the $classes map that execute() builds keyed by Content-ID. The batch API does not guarantee that parts come back in the order they were sent — that is why each part carries a Content-ID — so a reordered response was decoded into the wrong model class (or into a raw Response), and alt=media detection used the wrong request. This was a regression from the Guzzle 6 rewrite, which replaced the Content-ID lookup (`if (isset($classes[$key]))`) with positional matching and left $classes as an unused parameter. Resolve both the originating request and the expected class from the part's Content-ID, falling back to the previous positional behaviour when the Content-ID is not one we sent.
|
Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA). View this failed invocation of the CLA check for more information. For the most up to date status, view the checks section at the bottom of the pull request. |
cy-yun
left a comment
There was a problem hiding this comment.
This is a great fix for the regression introduced by the Guzzle 6 rewrite. Re-associating the batched responses to their requests via the Content-ID header is the correct approach as per the Batch API specification, since parts are not guaranteed to maintain order.
The tests thoroughly verify the logic. I've left a couple of minor inline suggestions to improve PHP 8.1+ compatibility and simplify the fallback logic, but overall the logic is sound!
Co-authored-by: Charlotte Y <38296042+cy-yun@users.noreply.github.com>
Co-authored-by: Charlotte Y <38296042+cy-yun@users.noreply.github.com>
bshaffer
left a comment
There was a problem hiding this comment.
This was a regression from the Guzzle 6 rewrite
What guzzle 6 rewrite?
Batch::parseResponse() paired each response part with a request by position ($requests[$i-1]), and silently ignored the $classes map that execute() builds keyed by Content-ID. The batch API does not guarantee that parts come back in the order they were sent — that is why each part carries a Content-ID — so a reordered response was decoded into the wrong model class (or into a raw Response), and alt=media detection used the wrong request.
This was a regression from the Guzzle 6 rewrite, which replaced the Content-ID lookup (
if (isset($classes[$key]))) with positional matching and left $classes as an unused parameter.Resolve both the originating request and the expected class from the part's Content-ID, falling back to the previous positional behaviour when the Content-ID is not one we sent.