[
https://issues.apache.org/jira/browse/THRIFT-908?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12914133#action_12914133
]
Christian Lavoie commented on THRIFT-908:
-----------------------------------------
So, after more digging down that path, it comes to a philosophical decision --
the cpp and java bindings skip fields if they're missing, badly typed or
otherwise unrecognizable; short version is that thrift-generated code does a
best effort attempt at reading as much as it can off the wire and ignores the
rest.
This patch changes that philosophy, and needs more discussion (after thinking
about it, I'm not sure I want it applied myself). Certainly not a candidate to
get applied before 0.5.
> Make required types actually required by the Haskell type system
> ----------------------------------------------------------------
>
> Key: THRIFT-908
> URL: https://issues.apache.org/jira/browse/THRIFT-908
> Project: Thrift
> Issue Type: Bug
> Components: Haskell - Compiler, Haskell - Library
> Environment: ghc 6.12.3 on Mac OS X 10.6 (Haskell Platform 2010.2)
> Darwin -- 10.4.0 Darwin Kernel Version 10.4.0: Fri Apr 23 18:28:53 PDT 2010;
> root:xnu-1504.7.4~1/RELEASE_I386 i386
> The Glorious Glasgow Haskell Compilation System, version 6.12.3
> Reporter: Christian Lavoie
> Attachments: v1-thrift-make-required-fields-required.patch
>
> Original Estimate: 6h
> Remaining Estimate: 6h
>
> Currently Haskell bindings consider all struct fields to be optional by
> having all struct fields be "Maybe $field". This patch removes that Maybe for
> required fields.
> THIS BREAKS EXISTING CODE.
> It's also probably incomplete -- I haven't looked too carefully yet at what
> happens on the server side if it tries to parse out an incomplete struct
> definition:
> {noformat}
> struct myStruct {
> 1: required byte foo,
> }
> {noformat}
> became after some code changes
> {noformat}
> struct myStruct {
> 1: optional byte foo,
> }
> {noformat}
> And code generated for the first version is asked to read a struct from the
> second version where foo is missing.
> I suspect after this patch the Haskell server commits hara-kiri trying to
> evaluate {{undefined}} _outside_ of the thrift generated code (so the client
> could would receive an error for something that the thrift generated code
> should have noticed as wrong). This is clearly inadequate.
> Consider this bug a thought experiment / request for comments until I
> validate that the patch is complete. It will still break existing code.
> First and foremost:
> # Thoughts?
> # What would be an acceptable way to handle the case described above for
> other Haskell users? Break the current interface some more by returning some
> error type (Either $struct $errMessags?). Throw a Haskell exception?
> # How do the other bindings handle parsing invalid data off the network?
> # The generator and generated code is fairly painful to read. Would the
> powers that be welcome a cleanup patch that moves as much of the
> t_hs_generator.cpp code to a boost::format-style system that uses template
> strings that look like Haskell code, and reformats the output to be indented,
> proper style Haskell code as much as possible?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
