Myujth

I'm noticing some behaviour that I can't explain in solr (version 7.5). I have two documents that each contain a field with a full path to a file.

doc1: 
 path: ["/home/kyle/filea.txt"]
,
doc2: 
 path: ["/home/kyle/file1.txt"]

• If I issue a query: path:filea.*, doc1 is correctly returned.


• If I issue a query: path:file1*, doc2 is correctly returned.


• If I issue a query: path:"file1.*", doc2 is correctly returned.


• If I issue a query: path:file1.*, doc2 is NOT returned.

I have the default TokenizerChain on the Index Analyzer and the Query Analyzer, and the field is multi-valued.

So my question: What is solr/lucene doing behind the scenes that causes the query for:

• 
  <string><number>.*

to not return the document I expect, when the other generic cases of:

• 
  <string>.*(no trailing number),


• 
  <string><number>* (no dot in query)


• 
  "<string><number>.*" (query in quotes)

all return what I think they should?

Your analyzer splits strings into tokens based on the the rules specified in UAX#29. The rules that are of interest here are WB6 - WB12. It will not split groups of letters at a period (ex. an abbreviation), or groups of digits (ex. a decimal number), but a letter followed by digit (or vice versa) will be split.

That is:

• "one.two" becomes one token: "one.two". In doc1, you get the token: "filea.txt"


• "1.2" becomes one token: "1.2"


• "one.2" becomes two tokens: "one" and "2". In doc2, you get the tokens: "file1" and "txt"


• "1.two" becomes two tokens: "1" and "two"

The other thing to understand, is that wildcard queries are not tokenized, so they will not find patterns that would, after analysis, would span two tokens or, in this case, characters that would be eliminated in tokenization.

So, your queries:

• 
  path:filea.* looks for "filea." as a prefix. It finds it because "filea.txt" is a token present in the index.


• 
  path:file1* looks for "file1" as a prefix. It finds it because ""file1" is a token in the index.


• 
  path:"file1.*" is a phrase query, and there are no wildcards in phrase queries. So, "file1.*" gets passed through analysis, which eliminates the punctuation, and becomes "file1", which it finds in the index.


• 
  path:file1.* looks for "file1." as a prefix. "file1" and "txt" are in the index, but "file1." is not, so it doesn't find anything.