Mercurial: Changelog

debugcommands: add debugwireproto command We currently don't have a low-level mechanism for sending arbitrary wire protocol commands. Having a generic and robust mechanism for sending wire protocol commands, examining wire data, etc would make it vastly easier to test the wire protocol and debug server operation. This is a problem I've wanted a solution for numerous times, especially recently as I've been hacking on a new version of the wire protocol. This commit establishes a `hg debugwireproto` command for sending data to a peer. The command invents a mini language for specifying actions to take. This will enable a lot of flexibility for issuing commands and testing variations for how commands are sent. Right now, we only support low-level raw sends and receives. These are probably the least valuable commands to intended users of this command. But they are the most useful commands to implement to bootstrap the feature (I've chosen to reimplement test-ssh-proto.t using this command to prove its usefulness). My eventual goal of `hg debugwireproto` is to allow calling wire protocol commands with a human-friendly interface. Essentially, people can type in a command name and arguments and `hg debugwireproto` will figure out how to send that on the wire. I'd love to eventually be able to save the server's raw response to a file. This would allow us to e.g. call "getbundle" wire protocol commands easily. test-ssh-proto.t has been updated to use the new command in lieu of piping directly to a server process. As part of the transition, test behavior improved. Before, we piped all request data to the server at once. Now, we have explicit control over the ordering of operations. e.g. we can send one command, receive its response, then send another command. This will allow us to more robustly test race conditions, buffering behavior, etc. There were some subtle changes in test behavior. For example, previous behavior would often send trailing newlines to the server. The new mechanism doesn't treat literal newlines specially and requires newlines be escaped in the payload. Because the new logging code is very low level, it is easy to introduce race conditions in tests. For example, the number of bytes returned by a read() may vary depending on load. This is why tests make heavy use of "readline" for consuming data: the result of that operation should be deterministic and not subject to race conditions. There are still some uses of "readavailable." However, those are only for reading from stderr. I was able to reproduce timing issues with my system under load when using "readavailable" globally. But if I "readline" to grab stdout, "readavailable" appears to work deterministically for stderr. I think this is because the server writes to stderr first. As long as the OS delivers writes to pipes in the same order they were made, this should work. If there are timing issues, we can introduce a mechanism to readline from stderr. Differential Revision: https://phab.mercurial-scm.org/D2392

debugcommands: add debugserve command `hg serve --stdio` requires the exact command argument form `hg -R <path> serve --stdio` for security reasons. An upcoming commit will need to start an SSH protocol server process with custom settings. This commit creates a `hg debugserve` command for starting servers with custom options. There are no security restrictions and we can add options here that aren't appropriate for built-in commands. We currently only support starting an SSH protocol server using the process's stdio file descriptors. The server supports logging its I/O activity to a file descriptor number passed as a command argument. Differential Revision: https://phab.mercurial-scm.org/D2464

wireprotoserver: support logging SSH server I/O to a file descriptor We will soon introduce a debug command and tests for low-level I/O behavior of the SSH wire protocol. To facilitate this, we need to instrument the SSH server so it can log its I/O as events occur. We teach the SSH server to convert its stdout and stderr file objects into file object proxies. We configure these proxies to log to a file descriptor whose file number is specified via a config option. The idea is to have a future debug command start the SSH server process with access to an extra file descriptor that can be used by the server process to log I/O. Monitoring only the write I/O will be more robust than monitoring both writes and reads from the client process because read operations are not deterministic. This will matter for tests that capture raw I/O activity. Differential Revision: https://phab.mercurial-scm.org/D2463

util: enable observing of util.bufferedinputpipe Our file object proxy is useful. But it doesn't capture all I/O. The "os" module offers low-level interfaces to various system calls. For example, os.read() exposes read(2) to read from a file descriptor. bufferedinputpipe is special in a few ways. First, it acts as a proxy of sorts around our [potentially proxied] file object. In addition, it uses os.read() to satisfy all I/O. This means that our observer doesn't see notifications for reads on this type. This is preventing us from properly instrumenting reads on ssh peers. This commit teaches bufferedinputpipe to be aware of our observed file objects. We do this by introducing a class variation that notifies our observer of os.read() events. Since read() and readline() bypass os.read(), we also teach this instance to notify the observer for buffered variations of these reads as well. We don't report them as actual read() and readline() calls because these methods are never called on the actual file object but rather a buffered version of it. We introduce bufferedinputpipe.__new__ to swap in the new class if the passed file object is a fileobjectproxy. This makes hooking up the observer automatic. And it is a zero cost abstraction for I/O operations on non-proxied file objects. Differential Revision: https://phab.mercurial-scm.org/D2404

util: add a file object proxy that can notify observers There are various places in Mercurial where we may want to instrument low-level I/O. The use cases I can think of all involve development-type activities like monitoring the raw bytes passing through a file (for testing and debugging), counting the number of I/O function calls (for performance monitoring), and changing the behavior of I/O function calls (e.g. simulating a failure) (to facilitate testing). This commit invents a mechanism to wrap a file object so we can observe activity on it. We have similar functionality in badserverext.py. But that's a test-only extension and is pretty specific to the HTTP server. I would like a mechanism in core that is sufficiently generic so it can be used by multiple consumers, including `hg debug*` commands. The added code consists of a proxy type for file objects. It is bound to an "observer," which receives callbacks whenever I/O methods are called. We also add an implementation of an observer that logs specific I/O events. This observer will be used in an upcoming commit to record low-level wire protocol activity. A helper function to convert a file object into an observed file object has also been implemented. I don't anticipate any critical functionality in core using these types. So I don't think explicit test coverage is worth implementing. Differential Revision: https://phab.mercurial-scm.org/D2462

wireprotoserver: ability to run an SSH server until an event is set It seems useful to be able to start an SSH protocol server that won't run forever and won't call sys.exit() when it stops. This could be used to facilitate intra-process testing of the SSH protocol, for example. We teach the server function to loop until a threading.Event is set and invent a new API to run the server until an event is set. It also won't sys.exit() afterwards. There aren't many callers of serve_forever(). So we could refactor them relatively easily. But I was lazy. threading.Event might be a bit heavyweight. An alternative would be a list whose only elements is changed. We can't use a simple scalar value like a bool or int because those types are immutable. Events are what you use in systems programming for this use case, so the use of threading.Event seems justified. Differential Revision: https://phab.mercurial-scm.org/D2461