Disallow tables with `s3://bucket` location (without trailing slash after bucket name)

[findepi]

This was missed in test coverage in #17848

We should also define whether we accept locations like s3://bucket.

HMS rejects this (so it's hard to test in the above PR's test setup), but Trino does not.
So one can use such a location with Glue, for example. Athena won’t read from such table, but we may.

If we want to support these, we need to have test coverage for that. Marking release-blocker since it should be part of #17848 which was a release-blocker itself.

If we do not want to support these, we need to reject them on Trino side, otherwise such locations may be used in real world, and then we cannot take this back.

For context: I believe such locations weren't supported before #15921, but haven't checked.

[findepi]

This needs to be tested without HMS.
Maybe we have similar to ones in 854c15f#diff-c6ac76ce5a102f34fe5cdca7794cff7404bf8ed13740ae7232c25ad6205a1245
but with Glue and real S3.

cc @findinpath @alexjo2144 @ebyhr

[findepi]

I am taking a stab at this.

[martint]

Removing the release-blocker tag, since this is just about adding a test. As long as the underlying bug is fixed, we will go ahead and release.

[findepi]

@martint there is a misunderstanding. The release-blocker tag is for this decision:

If we want to support these, we need to have test coverage for that. Marking release-blocker since it should be part of #17848 which was a release-blocker itself.

If we do not want to support these, we need to reject them on Trino side, otherwise such locations may be used in real world, and then we cannot take this back.

sorry for not making this clear earlier.

[findepi]

We need to make this decision ASAP, otherwise we can paint ourselves into a corner and have to support s3://bucket locations even if we decide we do not want them.

as shown in #17923 they are currently accepted and can be used to write data (but read currently fails).

cc @electrum @findinpath @alexjo2144 @losipiuk on how to best block locations like s3://bucket that are not interoperable and are partially working today.

[findepi]

i've renamed this ticket to reflect the intent. Once more apologies for making this blurred initially.

[findinpath]

Adding here a testing scenario to provide more context about the issue:

AWS Glue, opposed to HMS, accepts any path for a table:

trino> create table hive.default.findinpathtop1 (c varchar) with (format='textfile', external_location='s3://trino-findinpath');
CREATE TABLE
trino> show create table hive.default.findinpathtop1;
                      Create Table                       
---------------------------------------------------------
 CREATE TABLE hive.default.findinpathtop1 (              
    c varchar                                            
 )                                                       
 WITH (                                                  
    external_location = 's3://trino-findinpath', 
    format = 'TEXTFILE'                                  
 )     


trino> insert into hive.default.findinpathtop1 values 'hello', 'world';
INSERT: 2 rows
// the output does land on the s3 bucket on the top level (as expected)

trino> select * from hive.default.findinpathtop1;
Query 20230615_204101_00004_wiskm failed: Path is not absolute: s3://trino-findinpath
io.trino.spi.TrinoException: Path is not absolute: s3://trino-findinpath
	at io.trino.plugin.hive.BackgroundHiveSplitLoader$HiveSplitLoaderTask.process(BackgroundHiveSplitLoader.java:318)
	at io.trino.plugin.hive.util.ResumableTasks$1.run(ResumableTasks.java:38)
	at io.trino.$gen.Trino_dev____20230615_203305_2.run(Unknown Source)
	at io.airlift.concurrent.BoundedExecutor.drainQueue(BoundedExecutor.java:80)
	at java.base/java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1144)
	at java.base/java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:642)
	at java.base/java.lang.Thread.run(Thread.java:1589)
Caused by: java.lang.IllegalArgumentException: Path is not absolute: s3://findinpath-findinpath
at com.google.common.base.Preconditions.checkArgument(Preconditions.java:218)
	at io.trino.hdfs.s3.TrinoS3FileSystem.keyFromPath(TrinoS3FileSystem.java:1020)
	at io.trino.hdfs.s3.TrinoS3FileSystem.listObjects(TrinoS3FileSystem.java:837)

[findepi]

I checked on Trino ~417 that we were allowing locations like s3://bucket, INSERT worked, SELECT did not.

I did not check earlier versions yet. However, that version may not be widely deployed yet since it had issues like #17803 (with different failure mode), let me keep the release-blocker tag. I hope we can get this clarified soon.