Setup of the C-toolchain as well as scala shoulnd't be a problem so long as
these are all no-cost tools.
________________________________
From: Interrante, John A (GE Research, US) <john.interra...@ge.com>
Sent: Thursday, March 11, 2021 11:34 AM
To: dev@daffodil.apache.org <dev@daffodil.apache.org>
Subject: RE: Acceptance criteria for merging DFDL-to-C backend (runtime2-2202)?
Is that all? :-). I would add some criteria testing runtime2's conformance to
the DFDL 1.0 specification as well. Here goes...
1) Sufficient functionality to describe at least one example, of sufficient
message complexity to indicate that other similar "real" examples should be
possible
Yup. We have 7 schemas in
daffodil-test/src/test/resources/org/apache/daffodil/runtime2 that have "real"
examples of messages with sufficient complexity to indicate that other real
messages should be possible.
2) contains built-in tests for one or several such examples, showing each
supported aspect working.
Yup. We have TDML tests and binary data/XML infoset files for each of
these 7 schemas.
3) the tests are fully integrated - they run every time 'sbt test' is run on
daffodil.
Yup. We have Scala test classes for each of these 7 schemas' TDML
tests in daffodil-test/src/test/scala/org/apache/daffodil/runtime2.
4) setup instructions for developers are there so that people know how to
insure the required C tool-chain elements are there. These need to work on
Linux and Windows.
Needs work. The top-level README lists daffodil-runtime2's build
requirements under Build Requirements and has a single sentence under Getting
Started telling developers that they will need a C toolchain in order to build
daffodil-runtime2. However, I didn't make it clear in the README that
developers must setup a C toolchain as well as a Java/Scala toolchain and I
need to add a "C Setup and Notes" page to the Confluence Wiki with step-by-step
instructions showing developers how to setup a C toolchain on Linux and
Windows. I hope developers won't mind if we make it mandatory to setup both
Java/Scala and C toolchains before building Daffodil. I don't like the idea of
some developers building Daffodil without checking that the C files compile
successfully and the runtime2 tests pass successfully. I think it's sufficient
to guarantee that end-users of Daffodil never have to install a C toolchain to
use Daffodil. When end-users install Daffodil, they will get a pure Scala
Daffodil just like before. The only user-visible change is that this Daffodil
will know how to generate, compile, and run C code on the end-user's system if
the user asks Daffodil to do that.
I also would add another criteria:
5) A subset of Daffodil's DFDL 1.0 specification conformance TDML tests
modified to run on runtime2 as well as runtime1 during 'sbt test'.
We have lots and lots of TDML tests checking that Daffodil conforms to the DFDL
1.0 specification. We should be able to find a subset of these tests that
should pass on runtime2 as well and verify that they do keep passing when run
on both runtime1 and runtime2. This step would go a long way to ensure that
people working on other aspects of Daffodil do not break the C backend
inadvertently without knowing it. Note that this makes it even more mandatory
that Daffodil developers setup both Java/Scala and C toolchains. What do you
all think?
-----Original Message-----
From: Beckerle, Mike <mbecke...@owlcyberdefense.com>
Sent: Thursday, March 11, 2021 9:27 AM
To: dev@daffodil.apache.org
Subject: EXT: Re: Acceptance criteria for merging DFDL-to-C backend
(runtime2-2202)?
Acceptance criteria to me:
1) Sufficient functionality to describe at least one example, of sufficient
message complexity to indicate that other similar "real" examples should be
possible
2) contains built-in tests for one or several such examples, showing each
supported aspect working.
3) the tests are fully integrated - they run every time 'sbt test' is run on
daffodil.
4) setup instructions for developers are there so that people know how to
insure the required C tool-chain elements are there. These need to work on
Linux and Windows.
I believe these things insure that people working on other aspects of Daffodil
will not be inadvertently breaking the C backend indetectably. That's one major
concern I would have.
________________________________
From: Interrante, John A (GE Research, US) <john.interra...@ge.com>
Sent: Wednesday, March 10, 2021 4:24 PM
To: dev@daffodil.apache.org <dev@daffodil.apache.org>
Subject: Acceptance criteria for merging DFDL-to-C backend (runtime2-2202)?
Thanks to Mike's suggestion, I have moved the Runtime2 ToDos (changes requested
by reviewers) from the
DAFFODIL-2202<https://issues.apache.org/jira/browse/DAFFODIL-2202> issue to an
AsciiDoc document in the dev/design-notes subtree of the Daffodil website which
you can read here:
https://daffodil.apache.org/dev/design-notes/runtime2-todos/
I would like to discuss which acceptance criteria the runtime2-2202 development
branch must meet before I can submit a pull request to merge the DFDL-to-C
backend and code generator into the main branch. I plan to address the
Runtime2 ToDos and I especially want to run some of Daffodil's TDML tests on
the new runtime2 backend as well as the runtime1 backend by adding
defaultImplementations="daffodil daffodil-runtime2" to certain TDML tests'
attributes. (Although I suggest we use the shorter name "daf-c" or "runtime2"
because "daffodil daffodil-runtime2" is a lot of characters to put into the
defaultImplementations attribute.)
Daffodil's Confluence describes Runtime2's design here:
https://cwiki.apache.org/confluence/display/DAFFODIL/WIP:+Daffodil+Runtime+2
In particular, it suggests we divide the implementation of runtime2 into two
distinct phases:
* Phase 1 (aka Runtime2P1): No expressions. All lengths are fixed. All
arrays have fixed length.
* Phase 2 (aka Runtime2P2): Adding the DFDL expression language, lengthKind
'explicit', occursCountKind 'expression'.
I think phase 1 is almost done but we need to run a subset of Daffodil's TDML
tests on runtime2 before we can really say for sure.
Here is an initial set of discussion points - more questions and criteria are
welcome:
1. Which Daffodil TDML tests do we need to run on runtime2 to assert that
phase 1 is complete?
2. Can we merge runtime2 when these tests pass and then build out phase 2 in
the main branch, hopefully with help from other developers once they see how
useful phase 1 is?
3. Which of the Runtime2 ToDos need to be done before the merge as well?
Once we agree on a minimal set of acceptance criteria for the merge, I'll copy
the criteria to the JIRA issue.
