lucenenet-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Shad Storhaug <s...@shadstorhaug.com>
Subject RE: Remaining Work/Priorities
Date Thu, 15 Dec 2016 18:34:28 GMT
Update
======

It has been a while since I have communicated the current status of the Lucene.Net codebase
to the team, and I am getting concerned that claims that we are "close to release" are being
exaggerated a bit. We are almost ready to put a pre-release on NuGet so the masses can start
consuming it, but there are some bases we still need to cover to stabilize for release and
ready Lucene.Net 4.8.0 for enterprise-level quality expectations.

We have now successfully ported more than 380,000 executable lines of code from Java to .NET,
and have ported every Lucene sub-project that Itamar has earmarked as "important". We also
have support for .NET Core (at least on a branch: https://github.com/apache/lucenenet/pull/191)
and have over 6000 passing tests.

The following sub-projects (and their tests) that Itamar has earmarked as "optional" can still
be ported if any interested party wants to make a contribution. If not, they won't be in the
initial release.

1. Analysis.ICU
2. Analysis.Kuromoji (note only 3-4 days of work here, I think)
3. Analysis.Morfologik (Depends on Morfologik) 
4. Analysis.Phonetic (Depends on Apache Commons) Apache commons is mostly helper libraries,
so there's probably not real dependency just lots of replacement 
5. Analysis.SmartCN (note only 2-3 days of work here, I think)
6. Analysis.UIMA (Depends on Tagger, uimaj-core, WhiteSpaceTokenizer)
7. Demo (might be a good learning exercise) 

--------------------------------------------------------------------------------------------------------
Per Itamar: We should be ready to release and stamp our builds as 100% stable.
--------------------------------------------------------------------------------------------------------

Agreed. But we are not there, yet. There is still quite a bit of work to do on that front.

1. There are many issues with the public API of Lucene.Net.Core and a few other places (such
as Lucene.Net.Grouping). Most of these issues will require breaking API changes to fix . Although
most of these changes are fairly minor, since most will just be changing methods to properties
and vice versa, these changes will sweep through every consuming project. See API Phase 1
section below.
2. There are ~35 tests that are failing, and some others failing randomly, not to mention
some differences in failure counts between (https://github.com/apache/lucenenet/pull/191)
and master.
3. Some of the test framework is not yet complete, which explains some of the test failures.
The culture, time zone, and culture are not being randomized, we don't yet have the SuppressCodecs
functionality in place, nor has the Lucene 3.x backward compatibility been tested. The fact
that we are not randomizing culture means we are not testing the complete picture. We know
at present that there are ~35 test failures in en-US, but it is not currently known how many
we will have if we try other cultures. I suspect there are many issues around casing, date
and number formatting that will need to be addressed.
4. Bugs are still relatively easy to find in the codebase. For example, I recently discovered
that Atomaton doesn't fully support Unicode: https://github.com/apache/lucenenet/blob/master/src/Lucene.Net.Core/Util/Automaton/Automaton.cs#L648.
I also recently found and fixed an issue in the Test Framework where we were generating random
strings of numbers instead of random Unicode characters to create test strings, and discovered
some of the other random functions have bugs that make them not test the whole range. I could
go on, but the point is there are still many bugs that may negatively affect quality, most
which aren't causing obvious test failures.
5. We haven't had much feedback from end users, most likely because we haven't had many downloads
on MyGet, and because we haven't yet made the beta available on NuGet.

IMO we should do at least the following before we release (in order, bare minimum):

1. Fix the breaking API issues ASAP so we don't burden users with breaking changes later.
We can make all of these changes on #191 (or a branch of it) before it is merged so there
is only 1 build that breaks the API instead of doing it incrementally and having several successive
API-breaking packages.
2. Update documentation on GitHub home page (and possibly the website) so people can easily
report bugs, find the information about how to manually build, see that we have support for
NuGet and MyGet and where to get them, see current status, find documentation, contribute,
etc. Can we setup a WIKI on GitHub so we can add .NET specific Lucene documentation?
3. Ensure we use a version scheme we use allows for patching bugs so we aren't locked into
a single release of this port. Our current version scheme of MAJOR.MINOR.BUILD.REVISION-PRE
will work, but it is unclear whether the new build process supports this.
4. Get a wider beta release on MyGet (with icu-dotnet and .NET Core support) so we can start
getting user feedback on any remaining issues. We can then rely on the user feedback (or lack
thereof) as one factor to determine when we are ready for release.
5. Do a line-by-line sweep of Lucene.Net.Core and Lucene.Net.TestFramework (and possibly Lucene.Net.Classifications,
Lucene.Net.Expressions, Lucene.Net.Join) to ensure we have everything implemented and plugged
together correctly. The rest I am fairly confident is implemented correctly (we can rely on
user feedback for any potential issues).
6. Finish the Test Framework implementation so it randomizes culture, time zone, and codec
(and suppresses codecs correctly) so we get a true measure of test failures. And also so we
can determine if the Lucene.Net 3.x backward compatibility works.
7. Fix (at least the high priority) remaining tests.

Other tasks that are remaining to complete:

1. Create a RuleBasedBreakIterator based on icu-dotnet that breaks text similar to Java's
RuleBasedBreakIterator (for Highlighter and possibly Analysis.Th).
2. Fix the issues with Collator and RuleBasedCollator that are causing some test failures
(for Collation namespace in Analysis.Common).
3. Clean up the Support namespace, remove unused types, organize into sub-namespaces.
4. Make/port tests for types in the Support namespace to verify stability.
5. (recommended) Try to get others involved in the project to make high-level integration
APIs for certain target frameworks (see API Phase 2 below). This could be done after release,
but we might have more possibilities if we finish this part while it is a pre-release.
6. (optional) Performance tuning.
7. Suppress insignificant compiler warnings to see if there are any important ones left to
deal with (Lucene.Net.Core and a few others)
8. Finish XML documentation comments (Lucene.Net.Core, Lucene.Net.Classification, Lucene.Net.Queries)
9. Fix any directory casing issues in the codebase that can potentially cause problems on
some platforms (see https://github.com/apache/lucenenet/pull/196).

@Connie

1. Are there any remaining issues around the build/deployment we still need to resolve (versioning,
integration with TeamCity, etc)? 
2. Are we able to utilize our current versioning scheme (MAJOR.MINOR.BUILD.REVISION-PRE)?
I have verified that NuGet behaves correctly with this scheme, and IMO it makes sense to use
this scheme on a port such as this one so we have a way to patch without incrementing beyond
the semantic version of Lucene we are emulating. It looks like this versioning issue has been
a roadblock for fixing bugs in previous Lucene.Net ports.

API
===

--------------------------------------------------------------------------------------------------------
API Phase 1 - Stabilization
--------------------------------------------------------------------------------------------------------

Phase 1 takes care of finishing the breaking API changes that are necessary to get from where
we are to where we need to be for release. This is so the naming and other conventions are
consistent with .NET and/or Lucene, to eliminate casts that are currently required to use
certain functionality (such as Grouping), and to identify other parts of the API that could
be improved (either to make it more similar to Lucene or to make it more usable/intuitive
in .NET).

--------------------------------------------------------------------------------------------------------
Per Itamar: Public API Inconsistencies. We can discuss what should be done and what not when
we get to that stage. Some are an obvious "fixme" but some will break code compatibility with
Java I think we should avoid.
--------------------------------------------------------------------------------------------------------

We are now at the stage where we should make this a top priority. Mind sharing your thoughts
on what "needs to be compatible with Java"? It seems that MSDN has clear information on how
to differentiate between a property and a method: https://msdn.microsoft.com/en-us/library/ms229054(v=vs.100).aspx,
but which of the items below are you concerned about? Shouldn't we be more concerned with
making it compatible with .NET than with Java?

Here is that list of API issues again:

1.	Method names that are still camelCase
2.	Properties that should be methods (because they do a lot of processing or because they
are non-deterministic)
3.	Methods that should be properties
4.	.Size() vs .Size vs .Count – should generally all be .Count (or .Length) in .NET
5.	Interfaces should begin with “I”
6.	Classes should not begin with “I” followed by another singular capital letter (for
some reason some of them were named that way)
7.	.CharAt() should probably be this[]
8.	Generic types nested within generic types (which cause Visual Studio to crash when Intellisense
tries to read them)

We should add to that list:

1. Fix member accessibility to match that in Java (virtual by default, non-virtual if "sealed"
specified, etc.), so the intent of the original design can be realized.
2. Rename Tokenattributes namespace to TokenAttributes (and any other namespaces that don't
follow .NET conventions).
3. Rename enumerations that were named with a "_e" suffix back to their original name (we
can do this by de-nesting them from the class they are in so the name doesn't collide with
a property).
4. Find any public APIs that are using nullable enumerations and try to find another solution
(such as making an overload that doesn't accept the parameter and/or making a NOT_SET state
(with the enum default value of 0) in the enumeration).
5. Try to find a better replacement for Number than object (possibly by using different overloads
that accept different numeric types and keeping track of the type that was passed).
6. Fields should generally not be public in .NET - we have several that were named using Pascal
Case, but in general they should be made into properties that are Pascal Case that are either
auto-implemented or backed by fields that are camelCase.
7. Change the Collector abstract class to an interface so it will support covariance (required
by Grouping). We could alternatively back the Collector abstract class by an interface, but
I think that would be confusing since every place where the abstract class is used now will
need to be replaced with the interface anyway. 

There are probably some additional issues that will cause API breaks to fix, but this most
likely makes up the bulk of them.

--------------------------------------------------------------------------------------------------------
API Phase 2 - .NETify
--------------------------------------------------------------------------------------------------------

Phase 2 is to make the API more .NET-friendly.

There are several places where we can add overloads and extension methods to make features
of Lucene.Net act in concert with .NET better. For example, we could add extra overloads on
FSDirectory that take a path as a string so consumers don't have to new up a (probably pointless)
DirectoryInfo instance, which would make it act more like the FileStream object in .NET. Also,
as I previously mentioned to Itamar, there are several parts of Lucene's design that were
specifically aimed at using anonymous classes. We can probably find ways to simulate this
using some helper extension methods and/or fluent builders.

IMO Lucene.NET is useful, but its API is very low-level which makes using it challenging to
learn and integrate with modern .NET projects. Like many other packages that are available
on NuGet, it could use some integration packages with various frameworks within .NET to make
it easier to use. For example, StructureMap has integrations with MVC 5 and WebAPI: http://structuremap.github.io/integrations/.
Here are a few ideas for integrations we could make for Lucene.Net:

1. Lucene.Net.AspNet - integration with ASP.NET/MVC core (for plugging common Lucene.Net features
into the startup pipeline, etc).
2. Lucene.Net.AspNet.Suggest - UI integration with ASP.NET/MVC core, making Suggest functionality
into an HTML helper and/or view component that can be consumed/customized easily.
3. Lucene.Net.Linq - This already exists, but perhaps we should contact the author about bringing
in as part of the main repo, or perhaps helping out with our API effort: https://github.com/themotleyfool/Lucene.Net.Linq
4. Lucene.Net.EntityFramework - ?
5. Lucene.Net.MVC5 - integration with ASP.NET MVC 5.
6. Lucene.Net.WebApi - integration with WebApi.
7. Lucene.Net.Wpf - integration with WPF.
8. Lucene.Net.AspNet.Facet - UI integration with ASP.NET/MVC core, making faceted search functionality
into an HTML helper and/or view component, as well as any support needed to setup the index
for faceted search.

This is just a short list to get the ideas flowing. But we should really aim to make Lucene.Net
support easy to use with a wide range of frameworks across both the .NET Framework and .NET
Core stacks.

Just to be clear, the idea here is that we keep the Phase 1 API in place - that is, we have
an API that looks pretty much the same as Lucene, but build high-level APIs on top of it to
integrate many common use cases/configurations for integrating with these other frameworks.

For example, in ASP.NET Core, it is recommended to use a singleton instance for an IndexWriter
rather than opening and closing it all of the time - ideally, this could be done in a way
that is familiar to the ASP.NET Core startup configuration API.

        // This method gets called by the runtime. Use this method to add services to the
container.
        public void ConfigureServices(IServiceCollection services)
        {
            // Add framework services.
            services.AddApplicationInsightsTelemetry(Configuration);

            services.AddMvc();

            // Add a Lucene IndexWriter to the container (as a singleton)
            services.AddIndexWriter("~/the_index/", <other lucene-specific options>);
        }

That one line of code would potentially save everyone who uses a Lucene.Net IndexWriter in
combination with ASP.NET Core several hours of research and testing.

In the past, no such integrations existed with Lucene.Net, and as a result the project's success
has been limited and the project has always teetered on the edge of oblivion. IMO, bringing
the API to the users instead of making them come and find it would make Lucene.Net a much
more useful tool that is accessible to many more people, and make recruiting help for future
porting efforts easier. Furthermore, these integration packages could act as an adapter API
that doesn't need to change much from one Lucene.Net port to the next which will ease upgrading.

I am not alone in thinking that the API of Lucene.Net falls short of where it should be:

https://simplelucene.codeplex.com/documentation
https://ayende.com/blog/158914/lucene-net-is-ugly

So let's not let Lucene.Net fall short of expectations again. Instead, let's aim for making
Lucene.Net into the de-facto standard full-text search engine that is (mysteriously) missing
from the .NET framework.

Thoughts? Ideas?


Thanks,
Shad Storhaug (NightOwl888)


P. S. Itamar - can we get an update as to the status of the new website?




-----Original Message-----
From: Shad Storhaug 
Sent: Wednesday, October 12, 2016 2:10 AM
To: 'dev@lucenenet.apache.org'
Cc: 'Connie Yau'; 'cribs2@gmail.com'; 'itamar.synhershko@gmail.com'
Subject: RE: Remaining Work/Priorities

Update
======

I have just pushed some commits that fix several bugs in the Lucene.Net.Codecs project (all
452 tests pass most of the time, a few random failures) and fix all but 4 of the failing tests
in Lucene.Net.Core.


Fix for Test Context
-------------------------

For now, I have added method override stubs to each subclass in order to add the [Test] attribute,
so NUnit will run them in the correct context. I did that on all of the superclass tests except
for the ones in QueryParser (since Itamar mentioned he would be working in that area). Itamar,
you will probably need to follow suit to get all of the QP tests to pass - namely with the
QueryParserTestBase and TestQueryParser classes.

I have carefully put all of these changes into a single commit so it can be reverted easily,
if this solution doesn't happen to be compatible with xUnit: https://github.com/apache/lucenenet/commit/2a79edea6359e1ee1f83269cc7dc3ef2753ebf2c.
Hopefully that makes life easier for @conniey.

@Itamar, let me know when this is completed on your end so I can do a double revert and squash
the test stubs from QueryParser into an all-inclusive revert-able commit.

We can now correctly see how many tests we have in the core. Currently there are 2730 - it
seems we are still missing 720 tests, assuming they all were for something port-able.


Remaining Tests
---------------------

Next I plan to work on locating any tests that we have missed (starting in the core). It seems
these fall into several categories:

1. Tests that have not yet been ported.
2. Tests that have been partially ported that have not been added to the project.
3. Tests that have been ported, but are missing the [Test] attribute.
4. Tests in classes that have been ported that have been commented out (presumably because
at the time they were ported the dependencies did not yet exist).
5. Tests that have been Ignored in .NET that were not in Java.
6. Tests that have NUnit Assume.That() logic that depends on some non-existant JRE condition,
so they are not running in .NET.

I'll make a quick effort to get them to pass, but the main goal will be to ensure they all
can run and are included in the project. Just a heads up that the number of test failures
is likely to increase on this pass (but the number of bugs will likely decrease).


Failing Core Tests
-----------------------

I have looked into the remaining tests somewhat. There are 2 issues that I need some input
on to solve.


TestRamUsageEstimator.TestSanity()

Java Lucene uses a JRE-specific API to determine how much header size to add on each field.
This makes the estimates higher in Java. But more importantly, this test is failing because
the estimate for a real string instance is coming back as the same size as its shallow size
(16 bytes in this case) - it needs to be at least 1 byte more than that for the test to pass.
In Java (at least in a 64 bit environment), there are an extra 4 bytes being added for each
field.

Technically, there is a way to get these numbers from .NET, but it involves calling undocumented
APIs using pointers and will likely be different from one .NET version to the next (a bad
idea for a project that needs to support multiple .NET versions). The only solution I can
think of is to hard code in an extra 4 bytes for 64 bit (and most likely 2 bytes for 32 bit)
in order to make the numbers for the instances larger than their shallow size. I suppose the
alternative would be to either comment out the string test or change it to >= make it pass.
Thoughts? Alternatives?


TestNumericDocValuesUpdates.TestUpdateOldSegments()

I discovered what the issue is here (normally that is the hard part), but it seems that the
proper solution is going to be a major task. The NamedSPILoader (backed by SPIClassIterator)
in Java Lucene is used as a service locator to load classes throughout the project. In the
Codec abstract class, it is used to load up the codec for the context it is used in. However,
our port of the NamedSPILoader simply loads all of the classes from the current AppDomain
without any way to order them or override them.

The problem is that in Lucene, this was meant to be an extension point. And this particular
test (and probably many more of them) uses that extension point to change the codec to a Mock
from the test framework. This line from TestRuleSetupAndRestoreClassEnv pretty much sums up
what the issue is:

> Debug.Assert(Codec is Lucene42RWCodec, "fix your classpath to have 
> tests-framework.jar before lucene-core.jar");

Basically, it is using a configuration file to order the classes that are loaded so the test
mocks take priority over the built-in codecs.

Just fixing the test could be done by making the static NamedSPILoader variable in the Codec
class internal and swapping in a test double. However, that doesn't solve the bigger issue
that Lucene.Net is missing its extensibility for anyone who wants to write their own codec
(or tap into one of the other extensibility points). I guess the bigger question is how important
will it be for anyone to extend Lucene codecs or inject dependencies into Analyzer factories?
There doesn’t appear to be any more extensibility than that in Lucene 4.8.0, but that could
change in more recent or future versions of Lucene.


CI Builds
-----------

Not working. Can someone look into that please?


Thanks,
Shad Storhaug (NightOwl888)



-----Original Message-----
From: Shad Storhaug
Sent: Wednesday, October 5, 2016 8:23 PM
To: dev@lucenenet.apache.org
Cc: Connie Yau; 'cribs2@gmail.com'
Subject: RE: Remaining Work/Priorities

> Analysis.ICU (Depends on ICU4j) hopefully we can remove the ICU DLLs from the analysis.commons
module?

Just for clarification, these are two entirely different things in Java. Analysis.Common (Analysis.Collator
and Analysis.Th) depends on parts of Java:

import java.text.BreakIterator;
import java.text.Collator;
import java.text.ParseException;
import java.text.RuleBasedCollator;

Highlighter.PostingsHighlighter and Highlighter.VectorHighlight also depend on parts of Java:

import java.text.BreakIterator;
import java.text.CharacterIterator;

Analysis.ICU depends on a separate (icu4j) package:

import com.ibm.icu.text.Normalizer;
import com.ibm.icu.text.Normalizer2;
import com.ibm.icu.text.Transliterator;
import com.ibm.icu.text.Replaceable;
import com.ibm.icu.text.Transliterator;
import com.ibm.icu.text.UTF16;
import com.ibm.icu.text.UnicodeSet;
import com.ibm.icu.text.FilteredNormalizer2;
import com.ibm.icu.text.Collator;
import com.ibm.icu.text.RuleBasedCollator;
import com.ibm.icu.util.ULocale;
import com.ibm.icu.text.RawCollationKey;

That said, icu4j DOES have Collator and RuleBasedCollator classes, but it DOES NOT have a
BreakIterator or CharacterIterator class. It is unclear whether the Collator from icu4j would
work as a replacement for the one in core Java.

When I was digging through the JDK code, I noticed that BreakIterator and RuleBasedCollator
have a lot of common ICU dependencies there, so even if the RuleBasedCollator from icu4j is
compatible, it might make sense for us to port the one from Java anyway so we are dealing
with the same shared dependencies in Analysis.Common.

Once we port over the classes from the Java JDK, we will be able to eliminate our current
ICU4NET dependency (and the platform issues that come with it). That said, porting over those
pieces could take considerable work. In the interim it might make sense to make separate projects/NuGet
packages to isolate the areas that depend on BreakIterator, CharacterIterator, and RuleBasedCollator
so the rest can be released for wide/cross-platform use. Perhaps we can even make a basic
(scaled down) BreakIterator for Highlighter that breaks on spaces between words and punctuation
between sentences, which wouldn't work for Thai, but would work for most other languages.

Porting the (icu4j) package is another complete ball of yarn, we should take a look at (https://github.com/sillsdev/icu-dotnet)
to see if there is enough overlap there to power Analysis.ICU (offhand it looks as though
some classes are missing, though). It is a wrapper around the C library - it may be that we
just need to port more of it to get all of the pieces we need.

Speaking of Collation, @ChristopherHaws have you made any more progress on Analysis.Collation?
Were you able to determine if icu-dotnet's collator will make the tests pass?

> I'm on it QueryParser.Flexible

Great. The TimeZone probably just needs more research to work out how to utilize (in order
to implement the failing test). Also, FYI MSDN's recommendation (https://msdn.microsoft.com/en-us/library/system.timezone(v=vs.110).aspx)
is to use TimeZoneInfo rather than TimeZone (I noticed that several of the tests were recently
modified to use TimeZone rather than TimeZoneInfo).

As for the culture, in .NET I am pretty sure that we need to pass it as a parameter to another
overload of `QueryParser.Parse` rather than making it a property of QueryParser. But we can
deal with that in one step after you have finished porting.

--

Shad Storhaug (NightOwl888)

-----Original Message-----
From: itamar.synhershko@gmail.com [mailto:itamar.synhershko@gmail.com] On Behalf Of Itamar
Syn-Hershko
Sent: Wednesday, October 5, 2016 5:28 AM
To: dev@lucenenet.apache.org
Cc: Connie Yau
Subject: Re: Remaining Work/Priorities

Awesome, thanks for all the hard work Shad!

Our first priority should be fixing all remaining tests - in particular the one in Core. We
should be ready to release and stamp our builds as 100% stable. As you mentioned, this could
be an infrastructure issue - hopefully *Connie *can give a status update on her effort on
the switch to xUnit?

With regards to Modules, here's an updated breakdown based on your email + forgotten pieces
+ my comments:

*Ported:*
Lucene.Net (Core) - 15 failing / 1989 total Lucene.Net.Analysis.Common - 0 failing / 1445
total Lucene.Net.Classification - 0 failing / 9 total Lucene.Net.Expressions - 0 failing /
94 total Lucene.Net.Facet - (including #188 will be) 0 failing / 152 total Lucene.Net.Join
- 0 failing / 27 total Lucene.Net.Memory - 0 failing / 10 total Lucene.Net.Misc - 2 failing
/ 42 total Lucene.Net.Queries - 2 failing / 96 total Lucene.Net.QueryParser - 1 failing /
203 total Lucene.Net.Suggest - 0 failing / 142 total

We should do a second pass on the pieces we marked as ported, just to make sure the port is
full and we didn't leave anything behind :)

*Need to be ported:*
Highlighter (Depends on Collator (which is still being ported) and BreakIterator (which we
don't have a solution that works in .NET core yet)) Spatial (has 3rd party libraries that
need to be updates) Spatial4n (https://github.com/synhershko/Spatial4N) needs to be brought
up to speed with spatial4j, dependencies of which may cause some issues....
Codecs
Partially ported, mostly the tests weren't ported Grouping Not urgent, but provides nice functionality
that users will probably like

The only part with dependencies seems to be the spatial module - I will have a look there
soon if you don't get to that before I do.

*Can wait* - some modules are less frequently used, we should stabilize and release first
and then work on them based on demand Analysis.ICU (Depends on ICU4j) hopefully we can remove
the ICU DLLs from the analysis.commons module? I keep getting reports on some issues they
are causing Analysis.Kuromoji Analysis.Morfologik (Depends on Morfologik) Analysis.Phonetic
(Depends on Apache Commons) Apache commons is mostly helper libraries, so there's probably
not real dependency just lots of replacement Analysis.SmartCN Analysis.Stempel (currently
in progress) Analysis.UIMA (Depends on Tagger, uimaj-core, WhiteSpaceTokenizer) Demo while
important because can help newbies, we can do better by providing docs and real world examples.
I'm on it QueryParser.Flexible

*No need to port* - neither are needed in our context Benchmark (many dependencies) Replicator
(many dependencies) Sandbox (Depends on Apache Jakarta)

Once all modules are ported and all tests are passing, I think we should get two more items
fixed before an official release:

1. .NET Core support - I'm not clear on the status of it at the moment. We probably want to
have it in for the release.

2. Public API Inconsistencies. We can discuss what should be done and what not when we get
to that stage. Some are an obvious "fixme" but some will break code compatibility with Java
I think we should avoid.

One last note - *Wyatt*, do we know why there are no CI builds lately?

--

Itamar Syn-Hershko
http://code972.com | @synhershko <https://twitter.com/synhershko> Freelance Developer
& Consultant Lucene.NET committer and PMC member

On Sun, Oct 2, 2016 at 10:01 PM, Shad Storhaug <shad@shadstorhaug.com>
wrote:

> Hello,
>
> I just wanted to open this discussion to talk about the work remaining 
> to be done on Lucene.Net version 4.8.0. We are nearly there, but that 
> doesn't mean we don't still need help!
>
>
> FAILING TESTS
> -------------------
>
> We now have over 5000 passing tests and as soon as pull request #188 (
> https://github.com/apache/lucenenet/pull/188) is merged, by my count 
> we have only 20 (actual) failing tests. Here is the breakdown by project:
>
> Lucene.Net (Core) - 15 failing / 1989 total Lucene.Net.Analysis.Common
> - 0 failing / 1445 total Lucene.Net.Classification - 0 failing / 9 
> total Lucene.Net.Expressions - 0 failing / 94 total Lucene.Net.Facet - 
> (including #188 will be) 0 failing / 152 total Lucene.Net.Join - 0 
> failing / 27 total Lucene.Net.Memory - 0 failing / 10 total 
> Lucene.Net.Misc - 2 failing / 42 total Lucene.Net.Queries - 2 failing 
> / 96 total Lucene.Net.QueryParser - 1 failing / 203 total 
> Lucene.Net.Suggest - 0 failing / 142 total
>
> The reason why I said ACTUAL tests above is because I recently 
> discovered that many of the "failures" that are being reported are 
> false negatives (in fact, the VS2015 NUnit test runner shows there are
> 135 failing tests total and 902 tests total that don't belong to any 
> project). Most NUnit 2.6 test runners do not correctly run tests in 
> shared abstract classes with the correct context (test setup) to make 
> them pass. These out-of-context runs add several additional minutes to the test run.
>
> As an experiment, I upgraded to NUnit 3.4.1 and it helped the 
> situation somewhat - that is, it ran the tests in the correct context 
> and I was able to determine that we have more tests than the numbers 
> above and they are all succeeding. However, it also ran the tests in 
> an invalid context (that is, the context of the abstract class without 
> any setup) and some of them still showed as failures.
>
> I know @conniey is currently working on porting the tests over to xUnit.
> Hopefully, swapping test frameworks alone (or using some of the new 
> fancy test attributes) is enough to fix this issue. If not, we need to 
> find another solution - preferably one that can be applied to all of 
> the tests in abstract classes without too much effort or changing them 
> so they are too different from their Java counterpart.
>
> Remaining Pieces to Port
> ---------------------------------
>
> I took an inventory of the remaining pieces left to port a few days 
> ago and here is what that looks like (alphabetical order):
>
> 1. Analysis.ICU (Depends on ICU4j)
> 2. Analysis.Kuromoji
> 3. Analysis.Morfologik (Depends on Morfologik) 4. Analysis.Phonetic 
> (Depends on Apache Commons) 5. Analysis.SmartCN 6. Analysis.Stempel 
> (currently in progress) 7. Analysis.UIMA (Depends on Tagger, 
> uimaj-core, WhiteSpaceTokenizer) 8. Benchmark (many dependencies) 9.
> Demo 10. Highlighter (Depends on Collator (which is still being
> ported) and BreakIterator (which we don't have a solution that works 
> in .NET core yet)) 11. Replicator (many dependencies) 12. Sandbox 
> (Depends on Apache Jakarta) 13. Spatial (Already ported in #174 
> (https://github.com/apache/ lucenenet/pull/174), needs a recent 
> version of spatial4n) 14. QueryParser.Flexible
>
> Itamar, it would be helpful if you would be so kind as to organize 
> this list in terms of priority. It also couldn't hurt to update the 
> contributing documents 
> (https://github.com/apache/lucenenet/blob/master/CONTRIBUTING.md,
> and
> https://cwiki.apache.org/confluence/display/LUCENENET/Current+Status
> with the latest information so anyone who wants to help out knows the 
> current status.
>
> Of course, it is the known status of dependencies that we need 
> clarification on. Which of these dependencies is known to be ported?
> Which of them are ported but are not up to date? Which of them are 
> known not to be ported, and which of them are unknown?
>
>
> Public API Inconsistencies
> ---------------------------------
>
> One thing that I have had my eye on for a while now is the 
> .NETification/consistency of the core API (that is, in the Lucene.Net 
> project). There are several issues that I would like to address including:
>
>
> 1.       Method names that are still camelCase
>
> 2.       Properties that should be methods (because they do a lot of
> processing or because they are non-deterministic)
>
> 3.       Methods that should be properties
>
> 4.       .Size() vs .Size vs .Count - should generally all be .Count in
> .NET
>
> 5.       Interfaces should begin with "I"
>
> 6.       Classes should not begin with "I" followed by another capital
> letter (for some reason some of them were named that way)
>
> 7.       .CharAt() should probably be this[]
>
> 8.       Generic types nested within generic types (which cause Visual
> Studio to crash when Intellisense tries to read them)
>
> ... and so on. The only thing is these are all sweeping changes that 
> will affect everyone helping out on Lucene.Net and anyone who is 
> currently using the beta. So, I just wanted to gather some input on 
> when the most appropriate time to begin working on these sweeping changes would be?
>
>
> Thanks,
> Shad Storhaug (NightOwl888)
>
>
>
>
>
>
>
Mime
View raw message