summaryrefslogtreecommitdiffstats
path: root/src/main.rs
Commit message (Collapse)AuthorAgeFilesLines
* Allow passing a filename to dwvshRory Dudley2024-07-081-2/+29
| | | | | | The dwvsh binary may optionally take a filename as the last argument. Instead of spawning an interactive shell, it will instead run a shell program at the path specified.
* Option parsing and versionRory Dudley2024-07-031-0/+17
| | | | | | | Add some basic logic for parsing commandline arguments. Also, use build.rs to embed the program version (and git commit) during the compile step. The program currently accepts the '--version' commandline argument to print the version, then quit.
* Handle STDERR, in addition to STDOUTRory Dudley2024-05-191-1/+1
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This patch overhauls the reading and reciting of verses, such that the redirection of STDERR (in addition to STDOUT, which was already a feature), is now possible. Removed the 'stdout' argument from recite(), since it is no longer needed with how incantations function. A verse's couplet indicator is now a u8, instead of a bool, with certain values corresponding to types of couplets, for instance: ls | grep Ca | lolcat ^ ^ ^ | | 2: right side of a couplet | 3: both sides of a couplet 1: left side of a couplet Incantions are no longer hanlded in rune.rs, and the task macros have been removed. Now, a verse incants itself, matching on its own meter to determine how to handle the next verse. The following runes were added to help with handling STDERR: Write2 -> 2> WriteAll -> &> Addendum2 -> 2>> AddendumAll -> &>> The 'io' field in verse was changed from an Option<Rune>, to an array of Runes, since a single verse might have multiple IO operations. The following fields were added to Verse, to assist with handling STDERR: ip -> List of filenames to read into STDIN op -> List of filenames to send STDOUT to ep -> List of filenames to send STDERR to Keep track of channels when reading a poem. Channels are relating to IO operations. If channel is None, words get pushed to the verse's primary stanza (i.e. the verb or the clause). If a channel is selected, words are pushed to one of the aforementioned new fields in Verse. Read -> ip Write/Addedum -> op Write2/Addedum2 -> ep WriteAll/AddendumAll -> op and ep Notes: This commit also added tests for the new Runes.
* Handle aliases in read()Rory Dudley2024-04-041-1/+1
| | | | | | Instead of handling aliases in the recite() function, which requires two loops to handle properly with the current implementation, offload checking for aliases to the read() function.
* Use $PS1 for the promptRory Dudley2024-03-301-17/+25
| | | | | Instead of passing a hard-coded value for the prompt, use $PS1. The default is '|> ', set in dist/etc/dwvshrc.
* Replace env!("HOME") with env::var("HOME")Rory Dudley2024-03-301-2/+0
| | | | | | Replaced all (non-test) instances of env!("HOME") with env::var("HOME"). The env! macro should only be used in instances where the environment variable should be resolved during compile time.
* Add wrapper for global shell environmentRory Dudley2024-03-301-5/+6
| | | | | | | | Instead of having to pass around a bunch of different data structures for various shell functions, create the wrapper compose::Environment, which serves as a global shell state. It is configured via login/profile/rc scripts initially, but can of course be modified throughout the lifetime of the shell.
* Remove hard-coded PATHRory Dudley2024-03-281-20/+10
| | | | | | | | | | | | | | | | | | | | | | | | | | Use $PATH, instead of a hard-coded PATH from main(). This means that there is no longer a need to pass around PATH to repl()/recite()/path::refresh(), since path::refresh() can call env::var directly. Since the hard-coded paths were removed, there needs to be some way to define $PATH. When running the debug build, dwvsh will look in 'dist/etc/dwvshrc' for the initial environment setup. For the release target, dwvsh will look in '/etc/dwvshrc'. After the global rc file is sourced, dwvsh will try to source ~/.dwvshrc if it exists, so users can extend their environment without root access (assuming a release install). Notes: Throughout a lot of this program, we're calling `env!("HOME")`, in order to get the user's home directory. Technically, this is not correct. The env!() macro resolves environment variables during compile time, while env::var() gets environment variables for the running process (i.e. the shell). See https://users.rust-lang.org/t/env-vs-env-var/88119 for more info. In the near future, this will need to be addressed. Might be worth looking into what other shells do, though one idea I had was to invoke '/usr/bin/id', grab the user's ID, and use it to grab the rest of the info from /etc/passwd. This would be handled in an /etc/dwvlogin or /etc/dwvprofile most likely.
* Update docs for repl()readRory Dudley2024-03-281-2/+4
| | | | | Updated the docs for repl() in main.rs, to include the 'at_prompt' function parameter.
* Rename prefresh()Rory Dudley2024-03-261-2/+1
| | | | | | Rename the path::prefresh() function to path::refresh(). Calling convention should be to `use crate::path;` or `mod path;`, and then call path::refresh(...), for verbosity.
* Remove unecessary double quotesRory Dudley2024-03-261-1/+1
| | | | Don't need double quotes if just printing a newline.
* Remove debug lineRory Dudley2024-03-241-2/+0
| | | | Remove println!() in main that was used for debugging the parser output.
* read() and recite() overhaulRory Dudley2024-03-231-10/+19
| | | | | | | | | | | | | | | | | | | | | | | | | Rebuilt the LR parser (i.e. read()) from the ground up. This required that some changes be made to recite(), in order to accomodate the new data structures. These data structures were each split out into their own file, in order to make working with each component a bit easier. In addition to reworking the parts of the parser already present, some new features were also added, such as: - Support for strings (' and ") - Support for environment variables ($) - Support for interpreting tild as $HOME (~) - Support for sub-reading and sub-reciting (`) Notes: This is a huge commit that changes almost the entire program (main.rs is still the same, except for imports). Ideally, huge sweeping changes like this should not occur on the codebase, but since this is still pre-alpha, I guess this is acceptable. This is far from the end of patch set, however, as there is quite a lot of cleanup that needs to be done. For instance, checking for internal poems and environment variables should get split out to their own functions/macros. There is also some defunct code (that's commented out), that is unlikely to be useful in the future.
* Fix handling of SIGINTRory Dudley2024-02-291-4/+17
| | | | | | | | | | | | | | | Keep track of a new atomic variable: at_prompt, which is set to true just before blocking on io::stdin.read_line, and set to false just calling Poem::read. Additionally, for background tasks, there is a new ps macro called btask, which changes the process group of commands that are forked into the background, so that they don't receive SIGINT from the keyboard. Notes: Changing the process group on the Command is done via CommandExt. More details here: https://doc.rust-lang.org/std/os/unix/process/trait.CommandExt.html#tymethod.process_group
* Remove zombie slayer code from main replRory Dudley2024-02-281-10/+0
| | | | | Reaping is currently handled by a signal handler, so this code is no longer necessary.
* Remove custom errors and fix background forkingRory Dudley2024-02-271-12/+11
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | Removes the custom errors in src/recite/erro.rs, and replaces them with std::io::Errors throughout (recite(), incant_, macros). Fixed a bug with the way forking to the background is handled, where registering the signal handler in main for all processes would break couplets (i.e. pipes). Instead, this sets up a new signal handler each time a process is forked into the background. It uses a Vec<i32> to keep track of all the background processes. Notes: First off, there is some defunct code in the main repl loop, which is an example of killing zombie processes after each prompt. This should be removed, but I kept it in, just in case I go back to it for some reason. To be honest, I have no clue why this code works. In theory, I should have to remove the pid from the pids: Vec<i32> if waitpid returns a positive integer. However, when I tried this, it completely broke the program. ¯\_(ツ)_/¯ Also, it's worth noting that registering a signal handler with signal_hook::low_level::register, is somewhat costly, according to their docs. Given that this only occurs for background processes that are forked, however, I think it is acceptable. Finally, we never unregister the signal handler, so I'm not sure if that's still hanging out in memory somewhere or no.
* Cleanup recite(), custom errors, fixed forkingRory Dudley2024-02-261-4/+16
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | First off, moved the giant match statements out of recite(), and into macros in src/recite/ps.rs. There still needs to be two, since any verse using the 'couplet' meter will need to redirect its STDOUT. Now the recite() function returns a Result<(), Mishap>, which can be invoked when calling the incant_ functions. Custom errors were added in the form of 'Mishap''s. They are intended to be returned from the incant_ functions, in the event that something goes wrong with the Command::spawn() or Child::wait(). They each take a String, which should be the verb or stanza that was entered by the user. The incant_ functions separate the functionality of each type of meter from the recite() function. They return a Result<i32, Mishap>, where i32 is the exit code of the program that ran, and Mishap is a possible error. Before, the shell was cheating at forking a process to the background. It would actually spawn a thread to wait for that process to finish. Now, the program simply registers a handler for SIGCHLD, and uses libc's waitpid() function to reap the child process, and print some output to the user, indicating that it's finished. Notes: This was a huge patch which did some desperately needed cleanup of the recite() function. Moving forward, will need to add more documentation, and will probably scrap the custom errors, since this implementation is a little half-baked. It's worth looking into in the future, but we can probably live with io::Error's for the time being. Fixing forking was a pretty big deal, though. In Linux, and other u**x-like operating systems, parent processes need to reap their child processes, otherwise they become zombies. Previously, the dwvsh did this by spawning a separate thread to wait for child processes that were forked to the background. Now, we are registering a handle for SIGCHLD, which is a signal that gets sent to the parent when one of their children finishes, or is killed. Using waitpid(2), we can determine which process ended, and do something about it. In the case of a processes that was forked into the background, when it finished, waitpid(2) will return its PID. For foreground processes, it returns -1.
* Replace ctrlc with signal-hooksignalRory Dudley2024-02-241-7/+7
| | | | | | | | | | | | | | | | | | | | | | | | | | | | Replaced the 'ctrlc' crate with 'signal-hook' for handling of SIGINT. The 'signal_hook::low_level::register' function is actually unsafe. However, according to https://docs.rs/signal-hook/latest/signal_hook/low_level/fn.register.html, it is only unsafe in the case of multithreaded applications. There are some race conditions as well. For instance, it appears that even when we fork to a child process, SIGINT is captured on both that process, as well as the shell. Notes: The replacement was motivated by the fact that 'ctrlc' appears to use a separate thread to handle interrupts. This is evident if you run: ps aux | grep dwvsh USER PID %CPU %MEM VSZ RSS TTY STAT START COMMAND user pid 0.0 0.0 71500 3072 term Sl+ 20:08 target/debug/dwvsh Further reading in 'man ps' under 'PROCESS STATE CODES', reveals that 'l' is a process state referring to multithreaded applications. Given the nature of interupts, this seems unnecessary. The issue where SIGINT is captured by both the shell, and child process will have to be addressed.
* Reorganization and commentsRory Dudley2024-02-221-420/+21
| | | | | | | | | | | | Broke out the structs for a poem into their own file: src/recite.rs. Also put the 'prefresh' function into it's own file: src/recite/path.rs. Commented most of the parser code (including structs and helper methods related to parsing (i.e. Verse, Stanza, Meter, Poem)). Renamed any instance of the 'paths' variable to 'path'. Notes: The biggest task now is to cleanup Poem::recite. It has a ton of bogus error messages, and (seemingly) redundant code.
* Add back the 'cd' commandRory Dudley2024-02-201-0/+17
| | | | Add back change directory functionality into the new parser.
* Parsing improvementsRory Dudley2024-02-201-214/+153
| | | | | | | | | | | | | | Now the parser goes char by char, since special characters like '|' and '&' don't necessarily have to be whitespace seperated. Also added some VERY basic error detection for the parser (revolving around special chars). Notes: Even with the improvements to the parsing, this will likely get scrapped in favor of a cleaner approach. There are a lot of edge cases that are either difficult to handle with the current way things are, or just aren't being handled at all. The current implementation is also wont for better error detection and messages.
* Pipes, forks, and consecutive callsRory Dudley2024-02-201-90/+407
| | | | | | | | | | This adds some preliminary support for pipes (|), forks (&), and consecutive command calls (&&) to the shell. Notes: This branch is a huge WIP, and am only pushing it, cause it's late, and want to have my changes saved. A lot of cleanup and comments will be necessary moving forward.
* Only print command name in error messagesRory Dudley2024-02-191-2/+8
| | | | | Only print out the command name in error messages, rather than printing out the full path to the command.
* Path refresh refactor, comments, and error messagesevryloopRory Dudley2024-02-191-43/+102
| | | | | | | | | | | | | | | | | | The 'eval' function was renamed to 'repl'. The code to refresh the $PATH was moved into it's own function: 'prefresh', and is now being called in three locations: - At the beginning of 'repl', before the main loop - Inside the main loop, possibly during a path search if the command is not initially found - Inside the main loop, if the call to Command::spawn() throws an error, and the error is ErrorKind::NotFound Doc comments were added for each function, as well as a few more comments throughout that detail the program's control flow. The error messages in the main repl loop were cleaned up to have a more consistent style, and to provide more/better detail.
* Better handling of errors during the forkRory Dudley2024-02-191-3/+12
| | | | | | | | | | | Adds two additional error checks when the shell forks: 1. Checks for permission (+r, +x) 2. Checks if the file exists The first error may occur if the user does not have read access to the file, or if the file is not executable. The second error may occur if a file was removed from the $PATH, and the $PATH hasn't been refreshed yet.
* Refresh path only if command is not foundRory Dudley2024-02-191-19/+34
| | | | | | | | | | | | | | | | | | | This is a modified implementation of the 'refresh path on every iteration of the loop' idea. It instead, only refreshes the path if the command is not found. After the first refresh, if the command still is not found, it throws and error. Notes: This is probably the most sane solution, however, it has an issue. It can detect new files in the path just fine, but it cannot detect if a file was removed from the path. It may be prudent to expand the error handling when we fork, to see what kind of error the process is returning, and handle it more apropriately. Another solution may be to check the always check the existence of a file in the path before returning it from the match closure. This will overall slow down the REPL, however, since we'd now be making that check twice.
* Refresh paths every loopRory Dudley2024-02-191-6/+9
| | | | | | | | | | | | | Implements the path refresh at the start of each REPL loop. On this commit, it is printing out how long it needed to refresh all the paths in milliseconds. Notes: This may be an easier solution than using inotify. There is the obvious downside of a small delay each time we need to print the loop, but the highest I've seen so far is around 12 milliseconds, which seems acceptable. Using inotify as an alternative, it adds quite a few more dependencies, and some overhead in way of a watcher.
* Call programs from full and relative pathsRory Dudley2024-02-171-9/+15
| | | | | | | | | | Allow user to input a fullpath or relative path to a file that should be forked to. Notes: Currently, this does not check whether or not that file at the path specified is executable or not, but, if it isn't we will throw an 'Unable to fork' error, before printing the next prompt.
* Control flow in cdRory Dudley2024-02-171-2/+2
| | | | | Make the control flow in our cd implementation a bit easier to read/follow.
* EOF behaviorRory Dudley2024-02-171-0/+1
| | | | Print a newline before quitting when we receive an EOF.
* Update cdRory Dudley2024-02-171-4/+1
| | | | | Instead of printing an error if the path is not specified, simply 'cd' into the user's home directory instead.
* Change directoriesRory Dudley2024-02-161-0/+19
| | | | Added logic to change directories with 'cd'.
* Capture SIGINTRory Dudley2024-02-161-0/+12
| | | | | | Added logic to capture the interrupt signal. Using a rust crate called 'ctrlc' to do the heavy lifting. Program will simply reprint the prompt on a new line if the interrup signal is detected.
* Detect EOFRory Dudley2024-02-161-1/+8
| | | | Added logic to detect for an EOF (i.e. <C-d>).
* Initial commitRory Dudley2024-02-161-0/+86
An extremely miniamal shell. It is capable of forking processes, and passing arguments to them, but that's pretty much it. Notes: This is pretty much a prototype, to see how easily something like a shell could be implemented in Rust.