This commit removes the internal `Error` type in favor of exclusively
using the `failure::Error` type, simplifying error construction in a
number of locations and reducing the number of error types in play.
This commit fixes an issue where if a library is renamed via the `name`
key in the `[lib]` section of the manifest then `wasm-pack` would try to
generate bindings for an noexistent wasm-file, generating an error.
The fix was to internally use `cargo_metadata` more aggressively and
move around where this data is generated. This ended up refactoring a
few locations, but this should also bring improved error messages for
`cargo metadata` as well as caching the resulting data more aggressively
to avoid recalculating it too much.
Closes#339
This commit switches wasm-pack to using a global cache for all download
binaries, living typically in a user's home directory. The intention
here is to aovid creating a `bin` folder in all wasm-pack projects and
additionally share downloads between projects to ensure that you're
downloading a minimal number of binaries from the network.
Along the way the downloading code was restructured to support a global
cache, but everything should largely be as it was before!
Closes#292
This lets us leverage `cargo` for semver finding and then ensure that we get the
exact same version of the CLI that cargo selected. It also lets us support fuzzy
dependencies like "0.2" instead of exact dependencies like "0.2.21" again.
The standard library version of an `RwLock` may get poisoned if a thread
panics that holds a write lock. The `RwLock` from parking_lot [1] does
not get poisoned it instead released the lock on a panic. This allows us
to simplify the `ProgressOutput` API since it no longer returns any
errors. No more panics can occur on `ProgressOutput::drop()`. The
`Error` enum can thus be simplified as well because there is no need to
convert `PoisonErrors` anymore.
[1] https://github.com/Amanieu/parking_lot
This will fix the broken spinners that indicate steps which are
currently in progress. To achieve this the behavior of the progressbar
helper had to be adjusted.
The multibar does not work while a child process is run with
`std::process::Command`. Using multiple spinners without a multibar is
problematic as well because emitting warnings, infos and errors while a
single spinner is active will duplicate the spinners message. The
spinner will also absorb the last warning/error/info. Instead of
publishing warnings, errors and infos immediately they will now be
cached and only output when the current spinner finishes.
To make sure we can output all warnings to the user ProgressBars are no
longer exposed to the caller. Instead the active spinner will be
finished implicitly when a new spinner is allocated with the `message()`
function. Instead of relying on the `done()` function to be called the
progressbar now implements the `Drop` trait which will finish the last
spinner automatically.
Unfortunately, this introduces members that have to be mutable. To not
have to deal with mutable ProgressOutput all over the place the members
were put inside an RwLock. This allows us to use ProgressOutput from
inside multiple threads to emit warnings and errors.
This commit does a few things:
- Sets up the ability to log information based off log level
- Figures out where to log the file
- Starts logging information in the program
As this is a first pass the logging is good enough to know where things
went wrong, but we can expand on this in the future to log many many
things. We just now have an initial implementation. The log is written
out to a file that can be read by the user if they pass in -v, -vv, etc.
where each v is an extra level of logging.
Remove quicli from the code base. It's a fantastic library to get
started, but in order to implement logging as well as just maintaining
the library in general it was easier to remove it and continue work
without it. While we were going to remove it in 0.4 we found it easier
to remove now to implement logging.
Alex Crichton