You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@uima.apache.org by Marshall Schor <ms...@schor.com> on 2016/02/19 23:44:53 UTC
UIMA v3 progress, and PEAR / JCas work
Hi,
The experimental v3 redesign now runs all the core test cases (some were
modified a bit), except the PEAR test.
I also have to fix the compressed serialization form 6 to be compatible with V2
serialized forms (TBD). form 4 had a test for that already, and it's passing.
Handling JCas conversions and Pears feels kind of complex. I'm hoping to do
"common use cases" more or less automatically, with some enhanced reporting.
I've put down some general thoughts about this in a wiki page here:
*https://cwiki.apache.org/confluence/display/UIMA/MigrationPearJCas*
Cheers. -Marshall
Re: UIMA v3 progress, and PEAR / JCas work
Posted by Marshall Schor <ms...@schor.com>.
Making a quick pass thru the other projects now... to remove compile errors.
Most are trivial, or are test cases using internals that have changed.
-Marshall
On 2/22/2016 9:25 AM, Marshall Schor wrote:
> When I started this work, I took a copy of the existing uimaj code, which has
> lots of projects besides uimaj-core. I've only been working on the uimaj-core,
> and a bit on the uimaj-tools (to do a new version of JCasGen).
>
> If you check out the branch, you'll discover that many (most) of the other
> projects have compile errors. This is likely due to some API changes in
> uimaj-core, which need to be either "adapted" (via an adaption layer), or
> "fixed" in uimaj-core if possible, or documented, perhaps with some migration
> tooling.
>
> I haven't gotten to this yet. It would be good to make progress on this -
> helping hands welcome :-) I'm guessing there are several minor tweaks needed to
> fix most of this, along with perhaps some more difficult issues (yet to be
> identified).
>
>
> The other issue (for running, not for building) is that V2 JCasgen'd classes
> don't work. I think getting some kind of (semi-)automated tooling working here
> is important, so that's going to be my next focus. Ideally, I'd like to be able
> to take a test case (which we already have) which has 2 "pears", and run it, by
> discovering that the JCasGen'd classes are (a) convertable V2 -> v3 via a
> runtime automated process, and (b) they are "promotable" to the containing class
> loader, and then arranging to do the converting / promotion at the startup of
> the run when the pipeline is being prepared, and reporting and saving the
> results, so users could avoid this overhead the next time they run.
>
> -Marshall
>
> On 2/22/2016 6:48 AM, Richard Eckart de Castilho wrote:
>> Do you think it would be a good point now to see if existing code builds against the branch?
>> More specifically, should I try setting up a DKPro Core build against it or would I have to
>> expect serious breakage?
>>
>> DKPro Core uses largely the JCasGen Maven plugin for JCas classes, but we also have some
>> classes were we actually did some modifications (specifically: adding new methods).
>>
>> -- Richard
>>
>>> On 19.02.2016, at 23:44, Marshall Schor <ms...@schor.com> wrote:
>>>
>>> Hi,
>>>
>>> The experimental v3 redesign now runs all the core test cases (some were
>>> modified a bit), except the PEAR test.
>>>
>>> I also have to fix the compressed serialization form 6 to be compatible with V2
>>> serialized forms (TBD). form 4 had a test for that already, and it's passing.
>>>
>>> Handling JCas conversions and Pears feels kind of complex. I'm hoping to do
>>> "common use cases" more or less automatically, with some enhanced reporting.
>>> I've put down some general thoughts about this in a wiki page here:
>>>
>>> *https://cwiki.apache.org/confluence/display/UIMA/MigrationPearJCas*
>>>
>>> Cheers. -Marshall
>
Re: UIMA v3 progress, and PEAR / JCas work
Posted by Marshall Schor <ms...@schor.com>.
When I started this work, I took a copy of the existing uimaj code, which has
lots of projects besides uimaj-core. I've only been working on the uimaj-core,
and a bit on the uimaj-tools (to do a new version of JCasGen).
If you check out the branch, you'll discover that many (most) of the other
projects have compile errors. This is likely due to some API changes in
uimaj-core, which need to be either "adapted" (via an adaption layer), or
"fixed" in uimaj-core if possible, or documented, perhaps with some migration
tooling.
I haven't gotten to this yet. It would be good to make progress on this -
helping hands welcome :-) I'm guessing there are several minor tweaks needed to
fix most of this, along with perhaps some more difficult issues (yet to be
identified).
The other issue (for running, not for building) is that V2 JCasgen'd classes
don't work. I think getting some kind of (semi-)automated tooling working here
is important, so that's going to be my next focus. Ideally, I'd like to be able
to take a test case (which we already have) which has 2 "pears", and run it, by
discovering that the JCasGen'd classes are (a) convertable V2 -> v3 via a
runtime automated process, and (b) they are "promotable" to the containing class
loader, and then arranging to do the converting / promotion at the startup of
the run when the pipeline is being prepared, and reporting and saving the
results, so users could avoid this overhead the next time they run.
-Marshall
On 2/22/2016 6:48 AM, Richard Eckart de Castilho wrote:
> Do you think it would be a good point now to see if existing code builds against the branch?
> More specifically, should I try setting up a DKPro Core build against it or would I have to
> expect serious breakage?
>
> DKPro Core uses largely the JCasGen Maven plugin for JCas classes, but we also have some
> classes were we actually did some modifications (specifically: adding new methods).
>
> -- Richard
>
>> On 19.02.2016, at 23:44, Marshall Schor <ms...@schor.com> wrote:
>>
>> Hi,
>>
>> The experimental v3 redesign now runs all the core test cases (some were
>> modified a bit), except the PEAR test.
>>
>> I also have to fix the compressed serialization form 6 to be compatible with V2
>> serialized forms (TBD). form 4 had a test for that already, and it's passing.
>>
>> Handling JCas conversions and Pears feels kind of complex. I'm hoping to do
>> "common use cases" more or less automatically, with some enhanced reporting.
>> I've put down some general thoughts about this in a wiki page here:
>>
>> *https://cwiki.apache.org/confluence/display/UIMA/MigrationPearJCas*
>>
>> Cheers. -Marshall
>
Re: UIMA v3 progress, and PEAR / JCas work
Posted by Richard Eckart de Castilho <re...@apache.org>.
Do you think it would be a good point now to see if existing code builds against the branch?
More specifically, should I try setting up a DKPro Core build against it or would I have to
expect serious breakage?
DKPro Core uses largely the JCasGen Maven plugin for JCas classes, but we also have some
classes were we actually did some modifications (specifically: adding new methods).
-- Richard
> On 19.02.2016, at 23:44, Marshall Schor <ms...@schor.com> wrote:
>
> Hi,
>
> The experimental v3 redesign now runs all the core test cases (some were
> modified a bit), except the PEAR test.
>
> I also have to fix the compressed serialization form 6 to be compatible with V2
> serialized forms (TBD). form 4 had a test for that already, and it's passing.
>
> Handling JCas conversions and Pears feels kind of complex. I'm hoping to do
> "common use cases" more or less automatically, with some enhanced reporting.
> I've put down some general thoughts about this in a wiki page here:
>
> *https://cwiki.apache.org/confluence/display/UIMA/MigrationPearJCas*
>
> Cheers. -Marshall