http

Signed-off-by: Paul Merlin <paul@gradle.com>

Signed-off-by: Jonathan Leitschuh <Jonathan.Leitschuh@gmail.com>

Do not calcualte the hash for the same script multiple times.

The hash is now stored in the text resource itself so we don't need

an externall managed cache. The UriTextResource implementation has

been thoroughly optimized for local file URIs. Previously reading

those in was a major bottleneck due to copying every byte into 4(!)

different buffers.

Adding the HashCode to the TextResource made me aware of the fact

that the tooling API depended on these interfaces, although it should

never be reading scripts. It turns out that the script was attached to

the BuildLayout, where it didn't really belong. It is now only created

once the build is running and we are actually trying to load the settings

file. There were a few changes to types marked with @UsedByScanPlugin.

These were all checked against the scan plugin code base to ensure they

are safe. The reported data is now more accurate as it returns a `null`

value for a non-existing settings file instead of reporting a dummy path.

Signed-off-by: Sam Holmes <sbholmes@gmail.com>

This commit restricts the number of cases where we're going to perform HTTP retries. Instead of any error,

we will now retry when:

- a connection timeout occurs (client or server)

- the server responds an error code between 500 and 600

- response code is a too many requests error

Fixes #7850

Instead of including the original HTTP request URI, the bottommost

exception for failed HTTP requests now includes the URI of the last

redirect location, if any.

By default, IBM JDK 8 doesn't enable TLSv1.2, which results in

`fatal alert: protocol_version` when communicating with maven central.

This commit enables TLSv1.2 for HTTPS on IBM JDK 8.

These exceptions were handled incorrectly throughout the whole

codebase, usually rethrown without restoring the interrupt status

or discarded entirely. This means that the system would not stop

executing even though the user wanted it to. In some cases this

also left the system in an inconsistent state, leading to deadlocks.

The most notable changes include:

- UncheckedException.rethrow automatically restores the interrupt status

- AsyncDispatch is guaranteed to deliver its messages, even when interrupted

- ExecHandle cancels the started process if it is interrupted while waiting

- ExecHandle disconnects from the process' output before killing it

- The worker API cancels the started work items if it is interrupted

- ManagedExecutors shut down immediately if they are interrupted while stopping

- We no longer log exceptions caused by interruption to the console

- Interrupting our caches no longer leaks file locks

So that a proxy or MITM doesn't terminate the connections so easily.

Issue #6461

Signed-off-by: Jakub Strzyżewski <github@jeikobu.net>

Signed-off-by: Lukasz Sanek <lukasz.sanek@workday.com>