RFR: 8358688: HttpClient: Simplify file streaming in RequestPublishers.FilePublisher [v3]

[Jaikiran Pai]

On Wed, 11 Jun 2025 08:56:01 GMT, Volkan Yazici <vyazici at openjdk.org> wrote:

>> Hello Volkan,
>> 
>>> Earlier the regular file check was performed by FIS::new at subscription, hence, NSFE (yes, FIS::new throws NSFE when passed a not-regular file, e.g., a directory) was thrown at subscribe()
>> 
>> The `NoSuchFileException` is a NIO construct and resides in `java.nio.file` package. So the older `java.io.FileInputStream`'s constructor throwing a `NoSuchFileException` felt odd to me. So I tried this:
>> 
>> 
>> jshell> new FileInputStream(new File("/tmp"))
>> 
>> 
>> that throws a `FileNotFoundException` (which is understandable) and not a `NoSuchFileException`
>> 
>> 
>> |  Exception java.io.FileNotFoundException: /tmp (Is a directory)
>> |        at FileInputStream.open0 (Native Method)
>> |        at FileInputStream.open (FileInputStream.java:185)
>> |        at FileInputStream.<init> (FileInputStream.java:139)
>> |        at (#1:1)
>> 
>> 
>> Was it something else in that test failure which was throwing the `NoSuchFileException`?
>
> @jaikiran, your observation is correct. My explanation was worded poorly: I did not want to mean `NSFE` was thrown by `FIS::new`, instead I was stressing that _the exception_ was thrown at subscription-time. All in all, my point still stands: I could not move certain checks from `subscribe()` to `create()`, because of subscription- versus assembly-time difference in exception semantics. I will try to rephrase:
> 
> * In the old version,
>   * if the passed `Path` was associated with the default file system, and hence, supports `Path::toFile`, `FIS::new` of OIO was used and it throws `FNFE` of OIO.
>   * Otherwise, `Files::newIS` of NIO was used throwing `NSFE` of NIO.
>   * There was no `NSFE`-to-`FNFE` translation for the latter case.
>   * Both `FIS::new`/`Files::newIS` invocations were at subscription-time.
> * In the new version, only `Files::newIS` is used, at subscription-time, with (hopefully) sufficient `NSFE`-to-`FNFE` translation to preserve the old exception semantics.

Hello Volkan, what you have looks good to me. Daniel too explained to me yesterday why this proposed way is a better idea.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/25662#discussion_r2139714335