This way, Setup functions have access to the list of hosts that share the server block, and also, if needed for some reason, the index of the server block in the input
startup and shutdown commands should only be executed once per appearance in the Caddyfile (naturally meaning once per server block).
Notice that we support multiple occurrences of startup and shutdown in the same server block by building the callback array incrementally as we parse the Caddyfile, then we append all the callbacks all at once. Quite literally, the OncePerServerBlock function executes only once per server block!
If each server block had only one sync.Once then all directives would refer to it and only the first directive would be able to use it! So this commit changes it to a map of sync.Once instances, keyed by directive. So by creating a new map for every server block, each directive in that block can get its own sync.Once which is exactly what is needed. They won't step on each other this way.
Turns out having each server block share a single server.Config during initialization when the Setup functions are being called was a bad idea. Sure, startup and shutdown functions were only executed once, but they had no idea what their hostname or port was. So here we revert to the old way of doing things where Setup may be called multiple times per server block (once per host associated with the block, to be precise), but the Setup functions now know their host and port since the config belongs to exactly one virtualHost. To have something happen just once per server block, use OncePerServerBlock, a new function available on each Controller.
NewTestController now sets the site root to '.' to accomodate Windows. This introduced a failure on Linux because we join "." and an absolute path in /tmp/ and end up looking for the temp file in the wrong place. This change puts the temp file under the current working directory, which should resolve the issue.
Most of the Windows test failures are due to the path separator not being "/". The general approach I took here was to keep paths in "URL form" (ie using "/" separators) as much as possible, and only convert to native paths when we attempt to open a file. This will allow the most consistency between different host OS. For example, data structures that store paths still store them with "/" delimiters. Functions that accepted paths as input and return them as outputs still use "/".
There are still a few test failures that need to be sorted out.
- config/setup/TestRoot (I hear this has already been fixed by someone else)
- middleware/basicauth/TestBrowseTemplate and middleware/templates/Test (a line endings issue that I'm still working through)
* As value mutex was copied and therefore synchronization worked wrong
* It's pretty big structure with reference types, so copying create unnecessary
pressure on GC
Signed-off-by: Alexander Morozov <lk4d4@docker.com>
The errors and logs can now have log rolling if provided by the user.
The current customisable parameter of it are:
The maximal size of the file before rolling.
The maximal age/time of the file before rolling.
The number of backups to keep.
