6 Bugs and ideas

If you find a bug or misfeature, don’t hesitate to report it, by using M-x report-emacs-bug. The same applies to feature requests. It is best to discuss one thing at a time. If you find several unrelated bugs, please report them separately.

Below is a list of some known problems with Eshell version 2.4.2, which is the version included with Emacs 22.

Differentiate between aliases and functions

Allow for a Bash-compatible syntax, such as:

alias arg=blah
function arg () { blah $* }
Pcomplete sometimes gets stuck

You press TAB, but no completions appear, even though the directory has matching files. This behavior is rare.

grep python $<rpm -qa>’ doesn’t work, but using ‘*grep’ does

This happens because the grep Lisp function returns immediately, and then the asynchronous grep process expects to examine the temporary file, which has since been deleted.

Problem with C-r repeating text

If the text before point reads "./run", and you type C-r r u n, it will repeat the line for every character typed.

Backspace doesn’t scroll back after continuing (in smart mode)

Hitting space during a process invocation, such as make, will cause it to track the bottom of the output; but backspace no longer scrolls back.

It’s not possible to fully unload-feature Eshell
Menu support was removed, but never put back
If an interactive process is currently running, M-! doesn’t work
Use a timer instead of sleep-for when killing child processes
Piping to a Lisp function is not supported

Make it so that the Lisp command on the right of the pipe is repeatedly called with the input strings as arguments. This will require changing eshell-do-pipelines to handle non-process targets.

Input redirection is not supported

See the above entry.

Problem running less without arguments on Windows

The result in the Eshell buffer is:

Spawning child process: invalid argument

Also a new less buffer was created with nothing in it… (presumably this holds the output of less).

If less.exe is invoked from the Eshell command line, the expected output is written to the buffer.

Note that this happens on NT-Emacs 20.6.1 on Windows 2000. The term.el package and the supplied shell both use the cmdproxy program for running shells.

Implement ‘-r’, ‘-n’ and ‘-s’ switches for cp
mv dir file.tar’ does not remove directories

This is because the tar option –remove-files doesn’t do so. Should it be Eshell’s job?

Bind standard-output and standard-error

This would be so that if a Lisp function calls print, everything will happen as it should (albeit slowly).

When an extension module fails to load, ‘cd /’ gives a Lisp error
If a globbing pattern returns one match, should it be a list?
Make sure syntax table is correct in Eshell mode

So that M-DEL acts in a predictable manner, etc.

Allow all Eshell buffers to share the same history and list-dir
There is a problem with script commands that output to /dev/null

If a script file, somewhere in the middle, uses ‘> /dev/null’, output from all subsequent commands is swallowed.

Split up parsing of text after ‘$’ in esh-var.el

Make it similar to the way that esh-arg.el is structured. Then add parsing of ‘$[?\n]’.

After pressing M-RET, redisplay before running the next command
Argument predicates and modifiers should work anywhere in a path
/usr/local/src/editors/vim $ vi **/CVS(/)/Root(.)  Invalid regexp:
"Unmatched ( or \\("

With zsh, the glob above expands to all files named Root in directories named CVS.

Typing ‘echo ${locate locate}/binTAB’ results in a Lisp error

Perhaps it should interpolate all permutations, and make that the globbing result, since otherwise hitting return here will result in “(list of filenames)/bin”, which is never valuable. Thus, one could cat only C backup files by using ‘ls ${identity *.c}~’. In that case, having an alias command name glob for identity would be useful.

Once symbolic mode is supported for umask, implement chmod in Lisp
Create eshell-expand-file-name

This would use a data table to transform things such as ‘~+’, ‘...’, etc.

Abstract em-smart.el into smart-scroll.el

It only really needs: to be hooked onto the output filter and the pre-command hook, and to have the input-end and input-start markers. And to know whether the last output group was “successful.”

Allow for fully persisting the state of Eshell

This would include: variables, history, buffer, input, dir stack, etc.

Implement D as an argument predicate

It means that files beginning with a dot should be included in the glob match.

A comma in a predicate list should mean OR

At the moment, this is not supported.

(+ RET SPC TAB’ does not cause indent-according-to-mode to occur
Create eshell-auto-accumulate-list

This is a list of commands for which, if the user presses RET, the text is staged as the next Eshell command, rather than being sent to the current interactive process.

Display file and line number if an error occurs in a script
wait doesn’t work with process ids at the moment
Enable the direct-to-process input code in em-term.el
Problem with repeating ‘echo ${find /tmp}

With smart display active, if RET is held down, after a while it can’t keep up anymore and starts outputting blank lines. It only happens if an asynchronous process is involved…

I think the problem is that eshell-send-input is resetting the input target location, so that if the asynchronous process is not done by the time the next RET is received, the input processor thinks that the input is meant for the process; which, when smart display is enabled, will be the text of the last command line! That is a bug in itself.

In holding down RET while an asynchronous process is running, there will be a point in between termination of the process, and the running of eshell-post-command-hook, which would cause eshell-send-input to call eshell-copy-old-input, and then process that text as a command to be run after the process. Perhaps there should be a way of killing pending input between the death of the process, and the post-command-hook.

Allow for a more aggressive smart display mode

Perhaps toggled by a command, that makes each output block a smart display block.

Create more meta variables
$!

The reason for the failure of the last disk command, or the text of the last Lisp error.

$=

A special associate array, which can take references of the form ‘$=[REGEXP]’. It indexes into the directory ring.

Eshell scripts can’t execute in the background
Support zsh’s “Parameter Expansion” syntax, i.e., ‘${name:-val}
Create a mode eshell-browse

It would treat the Eshell buffer as an outline. Collapsing the outline hides all of the output text. Collapsing again would show only the first command run in each directory

Allow other revisions of a file to be referenced using ‘file{rev}

This would be expanded by eshell-expand-file-name (see above).

Print “You have new mail” when the “Mail” icon is turned on
Implement M-| for Eshell
Implement input redirection

If it’s a Lisp function, input redirection implies xargs (in a way…). If input redirection is added, also update the file-name-quote-list, and the delimiter list.

Allow ‘#<word arg>’ as a generic syntax

With the handling of word specified by an eshell-special-alist.

In eshell-eval-using-options, allow a :complete tag

It would be used to provide completion rules for that command. Then the macro will automagically define the completion function.

For eshell-command-on-region, apply redirections to the result

So that ‘+ > 'blah’ would cause the result of the + (using input from the current region) to be inserting into the symbol blah.

If an external command is being invoked, the input is sent as standard input, as if a ‘cat <region> |’ had been invoked.

If a Lisp command, or an alias, is invoked, then if the line has no newline characters, it is divided by whitespace and passed as arguments to the Lisp function. Otherwise, it is divided at the newline characters. Thus, invoking + on a series of numbers will add them; min would display the smallest figure, etc.

Write eshell-script-mode as a minor mode

It would provide syntax, abbrev, highlighting and indenting support like emacs-lisp-mode and shell-mode.

In the history mechanism, finish the Bash-style support

This means ‘!n’, ‘!#’, ‘!:%’, and ‘!:1-’ as separate from ‘!:1*’.

Support the -n command line option for history
Implement fc in Lisp
Specifying a frame as a redirection target should imply the currently active window’s buffer
Implement ‘>func-or-func-list

This would allow for an “output translators”, that take a function to modify output with, and a target. Devise a syntax that works well with pipes, and can accommodate multiple functions (i.e., ‘>'(upcase regexp-quote)’ or ‘>'upcase’).

Allow Eshell to read/write to/from standard input and output

This would be optional, rather than always using the Eshell buffer. This would allow it to be run from the command line (perhaps).

Write a help command

It would call subcommands with --help, or -h or /?, as appropriate.

Implement stty in Lisp
Support rc’s matching operator, e.g., ‘~ (list) regexp
Implement bg and fg as editors of eshell-process-list

Using bg on a process that is already in the background does nothing. Specifying redirection targets replaces (or adds) to the list current being used.

Have jobs print only the processes for the current shell
How can Eshell learn if a background process has requested input?
Make a customizable syntax table for redirects

This way, the user could change it to use rc syntax: ‘>[2=1]’.

Allow ‘$_[-1]’, which would indicate the last element of the array
Make ‘$x[*]’ equal to listing out the full contents of ‘x

Return them as a list, so that ‘$_[*]’ is all the arguments of the last command.

Copy ANSI code handling from term.el into em-term.el

Make it possible for the user to send char-by-char to the underlying process. Ultimately, I should be able to move away from using term.el altogether, since everything but the ANSI code handling is already part of Eshell. Then, things would work correctly on MS-Windows as well (which doesn’t have /bin/sh, although term.el tries to use it).

Make the shell spawning commands be visual

That is, make (su, bash, ssh, etc.) be part of eshell-visual-commands. The only exception is if the shell is being used to invoke a single command. Then, the behavior should be based on what that command is.

Create a smart viewing command named open

This would search for some way to open its argument (similar to opening a file in the Windows Explorer).

Alias read to be the same as open, only read-only
Write a tail command which uses view-file

It would move point to the end of the buffer, and then turns on auto-revert mode in that buffer at frequent intervals—and a head alias which assumes an upper limit of eshell-maximum-line-length characters per line.

Make dgrep load dired, mark everything, then invoke dired-do-search
Write mesh.c

This would run Emacs with the appropriate arguments to invoke Eshell only. That way, it could be listed as a login shell.

Use an intangible PS2 string for multi-line input prompts
Auto-detect when a command is visual, by checking TERMCAP usage
The first keypress after M-x watson triggers

eshell-send-input

Make / more electric

so that it automatically expands and corrects file names, beyond what the em-elecslash module is able to do. Or make file name completion for Pcomplete auto-expand ‘/u/i/stdTAB’ to ‘/usr/include/stdTAB’.

Write the pushd stack to disk along with last-dir-ring
Add options to eshell/cat which would allow it to sort and uniq
Implement wc in Lisp

Add support for counting sentences, paragraphs, pages, etc.

Once piping is added, implement sort and uniq in Lisp
Implement touch in Lisp
Implement comm in Lisp
Implement an epatch command in Lisp

This would call ediff-patch-file, or ediff-patch-buffer, depending on its argument.

Have an option such that ‘ls -l’ generates a dired buffer
Write a version of xargs based on command rewriting

That is, ‘find X | xargs Y’ would be indicated using ‘Y ${find X}’. Maybe eshell-do-pipelines could be changed to perform this on-thy-fly rewriting.

Write an alias for less that brings up a view-mode buffer

Such that the user can press SPC and DEL, and then q to return to Eshell. It would be equivalent to: ‘X > #<buffer Y>; view-buffer #<buffer Y>’.

Make eshell-mode as much a full citizen as shell-mode

Everywhere in Emacs where shell-mode is specially noticed, add eshell-mode there.

Permit the umask to be selectively set on a cp target
Problem using M-x eshell after using eshell-command

If the first thing that I do after entering Emacs is to run eshell-command and invoke ls, and then use M-x eshell, it doesn’t display anything.

M-RET during a long command (using smart display) doesn’t work

Since it keeps the cursor up where the command was invoked.