Commands that should NOT be saved in the history can be defined via the HISTORY_IGNORE environment variable (array). Currently, the following options are supported:

• "all": No commands will be saved until this option is removed again
• "whitespace": ignores commands with a leading whitespace character
• "no_such_command": ignores commands with exit status NO_SUCH_COMMAND
• "regex:xxx": where xxx is a regex (https://doc.rust-lang.org/regex), ignores all commands that match this regex

Examples:

# echo @HISTORY_IGNORE

# let HISTORY_IGNORE = [ all ] # saved
# let HISTORY_IGNORE = [ whitespace ] # saved
#  true # ignored
#  let HISTORY_IGNORE = [  ] # saved
#  let HISTORY_IGNORE = [ whitespace ] # ignored
# history
echo @HISTORY_IGNORE
let HISTORY_IGNORE = [ all ] # saved
let HISTORY_IGNORE = [ whitespace ] # saved
 let HISTORY_IGNORE = [  ] # saved


# let HISTORY_IGNORE = [ no_such_command ]
# true # saved
#  true # saved
# false # saved
# trulse # ignored


# let HISTORY_IGNORE = [ 'regex:.*' ] # behaves like 'all'
# true # ignored
#  true # ignored
# false # ignored
# trulse # ignored

Solution: I implemented it in a way that is open for future extensions, unlike bashs implementation ;)

Drawbacks We need to keep the compiled regexes in a vector, so we have a small heap-usage there. However, currently there is nothing we can do about that.

TODOs:

• unit & integration tests

Fixes: #503 #502 (partly, see https://github.com/redox-os/ion/issues/503#issuecomment-323079748)

State: WIP

Other: Also open to discussion if you think I could implement something differently, etc.

This PR removes hardcoded signal numbers and adds -1 to the PID when sending SIGCONT. #29

Tested on FreeBSD, Arch Linux and the Windows Subsystem for Linux. Redox might be even more broken though.

Problem:Globbing during shell expansion (see https://github.com/redox-os/ion/issues/258)

Solution: I added an extra param to the Normal token which designates it as a normal token that's globable or not. Changes introduced by this pull request:

• Globbing gets moved and probably (isn't) yet parsed 100% correctly. I'm just submitting this PR so I can get in touch with contributors and notify other people of my attempt at trying to fix this.

Drawbacks:

Wildcards during quotes don't parse correctly.

TODOs: [what is not done yet.]

Make sure wildcards in quotes get parsed as text (if this happened during the previous version).

Fixes: [what issues this fixes.] https://github.com/redox-os/ion/issues/258 State: [the state of this PR, e.g. WIP, ready, etc.] Ready for review Blocking/related: [issues or PRs blocking or being related to this issue.]

Other: The author isn't all that used to programming in Rust.

Problem: [describe the problem you try to solve with this PR.]

We allocate a lot

Solution: [describe carefully what you change by this PR.]

Use SmallVec and SmallString (based on SmallVec) to allow small vectors and strings to be stored on the stack.

Changes introduced by this pull request:

Some public API functions that took Strings take SmallStrings now. Also, it's now even faster on my machine.

• [...]
• [...]
• [...]

Drawbacks: [if any, describe the drawbacks of this pull request.]

TODOs: [what is not done yet.]

Clean up the code, unify the methods of creating a singleton smallvec (probably with a macro), move the typedefs to a types module.

Fixes: [what issues this fixes.]

State: [the state of this PR, e.g. WIP, ready, etc.]

WIP

Blocking/related: [issues or PRs blocking or being related to this issue.]

Other: [optional: for other relevant information that should be known or cannot be described in the other fields.]

The above template is not necessary for smaller PRs.

Several changes were necessary to allow Ion to use variables for tuning the shell. The History structure, for example, no longer needs to have a max_size variable because this information is already stored as a variable in the shell. This patch adds a few system variables such as CWD, HISTORY_FILE, HISTORY_FILE_ENABLED, HISTORY_SIZE, and DIRECTORY_STACK_SIZE. It also adds the capability for history being logged to a history file, named ion_history. What's currently lacking from this implementation is limiting the size of the history log and not logging commands loaded from files in history.

Problem: The previous version of pipelines::collect was a dragon's nest of bitflags and break <label>. It was difficult to extend for features such as #132 and #133.

Solution: pipelines::collect has been replaced by pipelines::Collector, which encapsulates the logic of pipelines::collect in a recursive descent parser. Additionally, instead of taking a mutable reference to an Option<&str>, pipelines::Collector::run returns a Result<...>.

Changes introduced by this pull request:

• Removed pipelines::collect
• Added pipelines::Collector and replaced instances of pipelines::collect with the logical equivalent for pipelines::Collector.

Drawbacks:

• The error messages are still of questionable quality. The main problem is being constrained by peg: the error messages have to be &strs according to the documentation, which is just mind boggling.
• There's still a good amount of complexity, cyclomatic complexity warning notwithstanding

TODOs:

• [x] Add tests for obscure redirects like &>> and ^>>

State: WIP just so that it has more time for review. Also see TODOs.

Other: In addition to the mentioned issues I'm trying to figure out what to tackle next. Replacing PEG is a must, but there's also statements.rs which needs a similar makeover.

Problem: Implement the exists builtin mentioned in #409

Solution: I copied over the test builtin implementation, removed the stuff we don't need in exists and implemented the necessary new logic.

Changes introduced by this pull request:

• Added exists builtin
• Implemented the following flags: -a, -b, -d, -f, --fn, and -s
• Implemented the STRING check
• Implemented unit tests for all functions
• Implemented integration tests

TODOs:

• Maybe: binary_is_in_path should use the same logic as the one that is used to spawn new processes
• test_binary_is_in_path could benefit from some more unit tests, however this requires a more complex/enhanced testing/-directory structure

Fixes: #409 (partly)

State: ready

Blocking/related: #409

Other:

The above template is not necessary for smaller PRs.

Changes introduced by this pull request:

• Modified Pipeline::stdin's type to be a union of either a filename or a string input
• Added support in pipelines::Collector for recognizing herestrings
• Added required functionality in Shell::execute_pipeline

TODOs:

• [ ] Outstanding Bug: running tr '[a-z]' '[A-Z]' <<< "foo" produces no output
• [x] Add doc comments for the peg::Input enum
• [x] Add regression tests in pipelines.rs and integration tests in examples/ for herestrings

Fixes: Closes #133

State: WIP; see TODOs.

Other: This PR ended up being slightly different than #200; instead of holding onto the string and writing to stdin once the command had been spawned, this takes a cue from the implementation of &|: write the herestring to a pipe, close the writer end, and pass the reader end to the first command in the pipeline.

Problem: execve and fork_and_exec share a large chunk of code

Solution: Remove the argument building and path resolution in fork_and_exec and call execve instead

Changes introduced by this pull request:

• There is no more repeated code between fork_and_exec and execve

Drawbacks: Ion forks even if the file doesn't exist, whereas before it wouldn't.

State: Ready

Add multiple redirections, that is, support for things like:

cmd <file1 <file2 ^>> err-log > output | cmd2

In current ion this doesn't work.

This change at current would be breaking as the statement:

tr 'x' 'y' | tr 'y' 'z' < file > file2

redirects file to the first tr command. With this change, we would redirect it to the second.

Overview

We treat a multiple redirection as it's own job. The only difference between these jobs and normal jobs is that they only read from their source(s) and then write to their sink(s).

Parsing

Alter Pipline contain a vector of PipeItems. A PipeItem consists of a Job, a Vec<Redirection> and a Vec<Input>. The above statement would be parsed into two items, cmd together with [file1, file2] as input and [err-log, output] as outputs, and cmd2 with no Inputs or Redirections

Command Generation and Redirection Setup

Added two two new variants of RefinedJob, Cat and Tee, Cat for concatenating multiple inputs, Tee for splitting multiple outputs. Alter the generate_commands function to return a Vec<(RefinedJob, JobKind, Vec<Redirection>, Vec<Input>)>, then condense redirect_input and redirect_output into one function that takes ownership and returns a new Vec<(RefinedJob, JobKind)>. It is here that we create the Tee and Cat jobs that we need and put them in the right spot in the pipeline.

Execution

Add exec_multi_in and exec_multi_out to handle the multiple redirections. These are fairly straightforward, they just read their sources and write to their sinks. I didn't add any extra paths to exec_job as the new RefinedJob kinds are always part of a pipeline.

Problem: Missing string methods

Solution: Add some of them

Changes introduced by this pull request:

• $basename: Print NAME without leading directories
• $extension: Print extension of NAME
• $filename: Print NAME without extension (if there is one)
• $parent: Print parent of NAME (if there is one)