[
https://issues.apache.org/jira/browse/LUCENE-8650?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Daniel Meehl updated LUCENE-8650:
---------------------------------
Description:
All (I think) TokenStream implementations set a "final offset" after calling
super.end() in their end() methods. ConcatenatingTokenStream fails to do this.
Because of this, it's final offset is not readable and DefaultIndexingChain in
turn fails to set the lastStartOffset properly. This results in problems with
indexing which can include unsearchable content or IllegalStateExceptions.
ConcatenatingTokenStream also fails to reset() properly. Specifically, it does
not set its currentSource and offsetIncrement back to 0. Because of this,
copyField directives (in the schema) do not work and content becomes
unsearchable.
I've created a few patches that illustrate the problem and then provide a fix.
The first patch enhances the TestConcatenatingTokensStream to check for
finalOffset, which as you can see ends up being 0.
I created the next patch separately because it includes extra classes used for
the testing that Lucene may or may not want to merge in. This patch adds an
integration test that loads some content into the 'text' field. The schema then
copies it to 'content' using a copyField directive. The test searches in the
content field for the loaded text and fails to find it even though the field
does contain the content. Flip the debug flag to see a nicer printout of the
response and what's in the index. Notice that the added class I alluded to is
KeywordTokenStream .This class had to be added because of another (ultimately
unrelated) problem: ConcatenatingTokenStream cannot concatenate Tokenziers.
This is because Tokenizer violates the contract put forth by
TokenStream.reset(). This separate problem warrants its own ticket, though.
However, ultimately KeywordTokenStream may be useful to others and could be
considered for adding to the repo.
The third patch finally fixes ConcatenatingTokenStream by storing and setting a
finalOffset as the last task in the end() method, and resetting currentSource,
offsetIncrement and finalOffset when reset() is called.
was:
All (I think) TokenStream implementations set a "final offset" after calling
super.end() in their end() methods. ConcatenatingTokenStream fails to do this.
Because of this, it's final offset is not readable and DefaultIndexingChain in
turn fails to set the lastStartOffset properly. This results in problems with
indexing which can include unsearchable content or IllegalStateExceptions.
ConcatenatingTokenStream also fails to reset() properly. Specifically, it does
not set its currentSource and offsetIncrement back to 0. Because of this,
copyField directives (in the schema) do not work and content becomes
unsearchable.
I've created a few patches that illustrate the problem and then provide a fix.
The first patch enhances the TestConcatenatingTokensStream to check for
finalOffset, which as you can see ends up being 0.
I created the next patch separately because it includes extra classes used for
the testing that Lucene may or may not want to merge in. This patch adds an
integration test that loads some content into the 'text' field. The schema then
copies it to 'content' using a copyField directive. The test searches in the
content field for the loaded text and fails to find it even though the field
does contain the content. Flip the debug flag to see a nicer printout of the
response and what's in the index. Notice that the added class I alluded to is
KeywordTokenStream .This class had to be added because of another (ultimately
unrelated) problem: ConcatenatingTokenStream cannot concatenate Tokenziers.
This is because Tokenizer violates the contract put forth by
TokenStream.reset(). This separate problem warrants its own ticket, though.
However, ultimately KeywordTokenStream may be useful to others and and could be
considered for adding to the repo.
The third patch finally fixes ConcatenatingTokenStream by storing and setting a
finalOffset as the last task in the end() method, and resetting currentSource,
offsetIncrement and finalOffset when reset() is called.
> ConcatenatingTokenStream does not end() nor reset() properly
> ------------------------------------------------------------
>
> Key: LUCENE-8650
> URL: https://issues.apache.org/jira/browse/LUCENE-8650
> Project: Lucene - Core
> Issue Type: Bug
> Components: modules/analysis
> Reporter: Daniel Meehl
> Priority: Major
> Attachments: LUCENE-8650-1.patch, LUCENE-8650-2.patch,
> LUCENE-8650-3.patch
>
>
> All (I think) TokenStream implementations set a "final offset" after calling
> super.end() in their end() methods. ConcatenatingTokenStream fails to do
> this. Because of this, it's final offset is not readable and
> DefaultIndexingChain in turn fails to set the lastStartOffset properly. This
> results in problems with indexing which can include unsearchable content or
> IllegalStateExceptions.
>
> ConcatenatingTokenStream also fails to reset() properly. Specifically, it
> does not set its currentSource and offsetIncrement back to 0. Because of
> this, copyField directives (in the schema) do not work and content becomes
> unsearchable.
> I've created a few patches that illustrate the problem and then provide a fix.
> The first patch enhances the TestConcatenatingTokensStream to check for
> finalOffset, which as you can see ends up being 0.
> I created the next patch separately because it includes extra classes used
> for the testing that Lucene may or may not want to merge in. This patch adds
> an integration test that loads some content into the 'text' field. The schema
> then copies it to 'content' using a copyField directive. The test searches in
> the content field for the loaded text and fails to find it even though the
> field does contain the content. Flip the debug flag to see a nicer printout
> of the response and what's in the index. Notice that the added class I
> alluded to is KeywordTokenStream .This class had to be added because of
> another (ultimately unrelated) problem: ConcatenatingTokenStream cannot
> concatenate Tokenziers. This is because Tokenizer violates the contract put
> forth by TokenStream.reset(). This separate problem warrants its own ticket,
> though. However, ultimately KeywordTokenStream may be useful to others and
> could be considered for adding to the repo.
> The third patch finally fixes ConcatenatingTokenStream by storing and setting
> a finalOffset as the last task in the end() method, and resetting
> currentSource, offsetIncrement and finalOffset when reset() is called.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
