jovanbulck / jsh Goto Github PK

todo --> change this everywhere in the (old) code.
Note the stdbool header is already included in jsh-common.h

Work in progress -> see jobs-on-background branch

todo list:

• stop waiting for forked child when it was interrupted by ^Z
• catch SIGCHILD signals to handle background child process completion and background processes wanting user input
• find out the unit of backgrounding / suspension in the jsh grammar: cmd (the whole pipeline!) is the unit of suspension: ^Z will be interpreted as EXIT_FAILURE
• parse the explicit backrounding & sign --> input is the unit of explicit backgrounding ?? What todo with thins like sleep 2 && echo done & ??
• create a linked list of current background processes, assign each one a jsh-job-id , a list of its all its OS process ids in the pipeline and a status (SUSPENDED, RUNNING_ON_BG, WANTING_INPUT, ...)
• create bg fg jobs built-ins that manipulate the background_processes list and send a SIGCONT signal to the child processes

Current situation:

$ ls
Downloads/
Desktop/
$ cd do            ## Press [Tab]: nothing happens
$ cd Do            ## Press [Tab]: code completion to cd Downloads/
$ cd Downloads/

Request: complete code disregarding case. For instance:

$ ls
Downloads/
Desktop/
$ cd do            ## Press [Tab]: code completion to cd Downloads/:
$ cd Downloads/

We should create a --fun-mode, for experimenting with cool stuff. For example:

• Whenever a command executed correctly, show a sl with jo-shell on it in two goods wagons.
• Whenever a command executed with an error code, spam the user with the error code, or something else

See https://github.com/blog/1547-release-your-software

IDEA: push the cwd on a stack on each call to cd and write a simple built_in to e.g. pop the top cwd and cd into it; have a user_defined number of directories to remember (eg via an environment variable?)

--> would be really usefull and probably also easy to implement...

(I think something familiar is done in bash and zsh.)

• apt
  Pressing Tab ↹ should yield:
  list search show install remove edit-sources update upgrade full-upgrade
• apt-get
  Pressing Tab ↹ should yield:
  update upgrade install remove autoremove purge source build-dep dist-upgrade dselect-upgrade clean autoclean check changelog download
• apt-cache
  Pressing Tab ↹ should yield:
  gencaches showpkg showsrc stats dump dumpavail unmet search show depends rdepends pkgnames dotty xvcg policy
• ifconfig
  Pressing Tab ↹ should yield the various active internet connections:
  eth0 lo wlan0
• zenity
  Pressing Tab ↹ should yield:
  --help-all --calendar --entry --error --info --file-selection --list --notification --progress --question --warning --scale --text-info --color-selection --password --forms
• systemctl
  Pressing Tab ↹ should yield:
  start stop restart reload status is-enabled enable disable help daemon-reload
• uname
  Pressing Tab ↹ should yield:
  --all --kernel-name --nodename --kernel-release --kernel-version --machine --processor --hardware-platform
• git checkout
  Pressing Tab ↹ should yield all available branches.

Normally, cd - gets you in the previous directory.

Expected behaviour:

$ pwd
  /etc/apt
$ cd /mnt/Data
$ pwd
  /mnt/Data
$ cd -
$ pwd
  /etc/apt

Actual behaviour:

$ pwd
  /etc/apt
$ cd /mnt/Data
$ pwd
  /mnt/Data
$ cd -
  [DEBUG: You entered: 'cd -']
  [DEBUG: readcmd: cmd 'cd -' expanded to 'cd -']
  [DEBUG: alias: input resolved to: 'cd -']
  jsh: cd: No such file or directory
  [DEBUG: built-in: executed 'cd']
  [DEBUG: freeing comd struct for 'cd']
  [DEBUG: parseexpr: expr evaluated with return value 1]
  [DEBUG: readcmd: freeing memory for: 'cd']
$ pwd
  /mnt/Data

minimal reproduction:

$ alias key val
$ unalias key
$ alias key anotherval   # exits successfully
$ alias                  # BUG: key is not in the alias table?

We should also include a 32 bit (Linux) build in the release. To do so, build a release-version on a 32 bit system with sudo make install and upload /usr/local/bin/jsh to the release page

There may be a possibility we can get the shell working on android too :).

TODO:

• Compile readline for android

has something to do with the "F", "T" uppercase elements in the beginning. I"ll look into it

malloc(arg) macro to auto check the return value of malloc calls

see also https://www.gnu.org/software/gnulib/manual/html_node/Safe-Allocation-Macros.html

Maybe write a wrapper macro for printdebug statements so that the pre-compiler only includes the statements when DEBUG=1

See https://en.wikipedia.org/wiki/Variadic_macro and http://stackoverflow.com/questions/679979/how-to-make-a-variadic-macro-variable-number-of-arguments

• in makefile if check for sed mac version
• in installer script gmake instead of make
• no update-db in make

This is a nasty bug that took me some time ;-)

The problem is the OS limits the size of a pipe buffer and jsh doesn't start a built_in-producer in parallel with its consumer-follower in the pipeline (since jsh is mono-threaded).

context: jsh pipeline implementation

To clarify: let's take an easy non-built_in example cat some_file | grep some_string. This is more or less what happens:

• jsh parses the string and construct the pipeline of comd's in its data structures
• jsh then opens the necessary pipe buffers using libc functions (OS system calls)
• jsh then starts a loop where it forks all the commands in the pipeline on after the other, redirecting stdin and stdout to the correct pipe file descriptors just before calling execv to replace the forked jsh child with the correct UNIX command (e.g. cat or grep)
• After this loop, jsh waits for the completion of all children

The point is jsh doesn't block after one process in the pipeline is started (forked). Instead it directly creates the next one. This implies the consumer process (eg grep) can start consuming data from the pipe filled by the producer (eg cat). The OS limits the size of a pipe buffer, refusing any more data when the pipe buffer is full. The rationale here is that the consumer should consume some data, freeing up space, before the producer can continue producing.

problem: built_ins in pipelines

Now the problem: jsh deals with built_ins in a pipeline different then with normal "forkable" commands. For example consider history | grep some_string

• jsh starts the actual pipe processes in the fork loop above; when it detects one of the processes is a built_in, it first redirect stdin and stdout to the pipes and then calls the parse_built_in() function, blocking the loop till completion...

Of course this is the problem: a built_in will write all of its data to the pipe before the consumer process is started (since jsh executes in a single thread). When the history file is very large, the pipe buffer fills up... (e.g. + 60.000 bytes of history; +5000 lines)

possible solutions

Some solution sketches:

• a straightforward and very bad solution I think: make the program execute a built_in in a separate thread --> no-go
• hackhackhack: replace the history built_in with an external process that reads ~/.jsh_history e.g. simple cat --> this fixes the problem, by hacking around it; no real general solution
• the only good thing to do: find a way to execute a built_in and continue the loop
• e.g. by first executing the entire loop and remembering to execute built_ins after all processes are forked? I think this should work since the forked processes wait for an EOF on stdout anyway?

Comments? Feedback? Ideas?

If anyone feels like it, claim! This is a nice bugfix for getting to know the jhs internals :-)

We should have a bitcoin address, to fund continued development of jsh.

Request: implement .po translation files

Current behaviour:
Suppose the following alias is defined: ll='ls -la'.
Suppose ~/.jshrc contains the instruction: alias ll "ls -l"

$ alias
alias ll="ls -la"
$ source ~/.jshrc
$ alias
alias ll="ls -la"
alias ll="ls-l"

According to POSIX, in the case of a double aliasdefinition, the older alias should be overwritten with the newer one.

Expected behaviour:

$ alias
alias ll="ls -la"
$ source ~/.jshrc
$ alias
alias ll="ls -l"

just a test -- ignore

todo

• no error for config files that doesn't exist --> create new empty files && silently (debugoutput) ignore if fail
• no "errors will be printed like this"

--> push to master

For elegance and efficiency, we should probably change the datastructure for aliases from a linked list of structs to a hashmap<char *key, char *value>

As far as I know, there's no native support in libc for hashmaps. We should probably look to an optimized GPL licensed hashmap implementation and include it in the project...

Advanced users know how to use a Makefile, rookies don't. We should create an executable that installs jsh. The executable should use dialog for ease of use. You can find some images of what's possible with dialog right here.

Our installer should do the following:

• Tell the user what the installer is going to do, ask whether he wants to continue installing
• Ask the user where to install jsh
• Ask the user to create a symlink in /usr/bin
• Recognise the current distribution, and subsequently:
  • If the distro is known by jsh:
    • Check which libraries are already installed
    • Download and install the needed libraries via the system package manager (apt-get, pacman, tce-load)
  • If the distro is not known by jsh: ask the user to search for the needed libraries himself before continuing installation.
• Edit boot flags via the checklist module
• Look for a jsh config files in $HOME; if they are not there, ask the user if he wants the installer to create default jsh config files. The user should be able to edit the config files via the editbox module.
• Show make output with the progressbox module, or show the installation progress via the gauge or mixedgauge module
• The last step should be to ask the user whether he wants to set jsh as the default shell (with the chsh command)
• Notify the user that the installation completed succesfully, and exit cleanly.
• When the user repeatedly presses Enter, the installer should install with default idiot-proof settings.
• We this is done, ewe can divide the 'Installation' part on the wiki in 3 parts:
  1. Downloading a pre-built binary chmoding it and copy to /usr/bin/ if wanted
  2. 'Easy way of installing' using the dialog-based installer: walk trough with screen shots of every step
  3. Building jsh for your own system, using make
• Testing on various platforms, making the sh script as much portable as possible

Alternative: we can also use zenity. Not my weapon of choice, but we should consider it.

Currently, jsh is not on AUR!

Current behaviour

$ yaourt -Ss jsh
community/jshon 20131105-1
    A json parser for the shell.
aur/heirloom-sh-cvs 2011.06.22-1 (5)
    Traditional sh [jsh(1)], derived from OpenSolaris code. CVS repository
aur/heirloom-sh-musl-cvs 20120609-1 (1)
    sh [jsh(1)], derived OpenSolaris code, statically linked to musl libc
aur/nodejs-jshint 2.5.6-1 (33)
    Static analysis tool for JavaScript
aur/vim-lint-git 20130427-1 (1)
    Vim + JSHint + CSSLint + V8

Expected behaviour

$ yaourt -Ss jsh
community/jshon 20131105-1
    A json parser for the shell.
aur/jsh 1.1.0 [installed]
    A basic UNIX shell implementation in C
aur/heirloom-sh-cvs 2011.06.22-1 (5)
    Traditional sh [jsh(1)], derived from OpenSolaris code. CVS repository
aur/heirloom-sh-musl-cvs 20120609-1 (1)
    sh [jsh(1)], derived OpenSolaris code, statically linked to musl libc
aur/nodejs-jshint 2.5.6-1 (33)
    Static analysis tool for JavaScript
aur/vim-lint-git 20130427-1 (1)
    Vim + JSHint + CSSLint + V8 

e.g. !! will be expanded to the previous entered command

use the GNU history library?
--> see http://tiswww.case.edu/php/chet/readline/history.html#SEC3

installer.sh should be executable from the get-go. Current situation:

[gijs@therion /tmp]$ git clone https://github.com/jovanbulck/jo-shell.git
Cloning into 'jo-shell'...
remote: Counting objects: 762, done.
remote: Compressing objects: 100% (57/57), done.
remote: Total 762 (delta 26), reused 0 (delta 0)
Receiving objects: 100% (762/762), 456.96 KiB | 357.00 KiB/s, done.
Resolving deltas: 100% (456/456), done.
Checking connectivity... done.
[gijs@therion /tmp]$ cd jo-shell 
[gijs@therion jo-shell]$ ls -hl
totaal 180K
-rw-rw-r-- 1 gijs gijs 8,0K dec  6 11:36 alias.c
-rw-rw-r-- 1 gijs gijs 1,1K dec  6 11:36 alias.h
-rw-rw-r-- 1 gijs gijs  14K dec  6 11:36 installer.sh
-rw-rw-r-- 1 gijs gijs  390 dec  6 11:36 jo-shell.txt
-rw-rw-r-- 1 gijs gijs  26K dec  6 11:36 jsh.c
-rw-rw-r-- 1 gijs gijs 2,4K dec  6 11:36 jsh-colors.h
-rw-rw-r-- 1 gijs gijs 7,1K dec  6 11:36 jsh-common.c
-rw-rw-r-- 1 gijs gijs 3,6K dec  6 11:36 jsh-common.h
-rw-rw-r-- 1 gijs gijs 9,3K dec  6 11:36 jsh-completion.c
-rw-rw-r-- 1 gijs gijs 1,7K dec  6 11:36 jsh-completion.h
-rw-rw-r-- 1 gijs gijs 3,5K dec  6 11:36 jsh-man.1
-rw-rw-r-- 1 gijs gijs  21K dec  6 11:36 jsh-parse.c
-rw-rw-r-- 1 gijs gijs 2,2K dec  6 11:36 jsh-parse.h
-rw-rw-r-- 1 gijs gijs  35K dec  6 11:36 LICENSE
-rw-rw-r-- 1 gijs gijs 6,8K dec  6 11:36 Makefile
-rw-rw-r-- 1 gijs gijs 4,3K dec  6 11:36 mini-grep.c
-rw-rw-r-- 1 gijs gijs 3,4K dec  6 11:36 README.md

Desired situation:

[gijs@therion /tmp]$ git clone https://github.com/jovanbulck/jo-shell.git
Cloning into 'jo-shell'...
remote: Counting objects: 762, done.
remote: Compressing objects: 100% (57/57), done.
remote: Total 762 (delta 26), reused 0 (delta 0)
Receiving objects: 100% (762/762), 456.96 KiB | 357.00 KiB/s, done.
Resolving deltas: 100% (456/456), done.
Checking connectivity... done.
[gijs@therion /tmp]$ cd jo-shell 
[gijs@therion jo-shell]$ ls -hl
totaal 180K
-rw-rw-r-- 1 gijs gijs 8,0K dec  6 11:36 alias.c
-rw-rw-r-- 1 gijs gijs 1,1K dec  6 11:36 alias.h
-rwxrwxr-x 1 gijs gijs  14K dec  6 11:36 installer.sh
-rw-rw-r-- 1 gijs gijs  390 dec  6 11:36 jo-shell.txt
-rw-rw-r-- 1 gijs gijs  26K dec  6 11:36 jsh.c
-rw-rw-r-- 1 gijs gijs 2,4K dec  6 11:36 jsh-colors.h
-rw-rw-r-- 1 gijs gijs 7,1K dec  6 11:36 jsh-common.c
-rw-rw-r-- 1 gijs gijs 3,6K dec  6 11:36 jsh-common.h
-rw-rw-r-- 1 gijs gijs 9,3K dec  6 11:36 jsh-completion.c
-rw-rw-r-- 1 gijs gijs 1,7K dec  6 11:36 jsh-completion.h
-rw-rw-r-- 1 gijs gijs 3,5K dec  6 11:36 jsh-man.1
-rw-rw-r-- 1 gijs gijs  21K dec  6 11:36 jsh-parse.c
-rw-rw-r-- 1 gijs gijs 2,2K dec  6 11:36 jsh-parse.h
-rw-rw-r-- 1 gijs gijs  35K dec  6 11:36 LICENSE
-rw-rw-r-- 1 gijs gijs 6,8K dec  6 11:36 Makefile
-rw-rw-r-- 1 gijs gijs 4,3K dec  6 11:36 mini-grep.c
-rw-rw-r-- 1 gijs gijs 3,4K dec  6 11:36 README.md

When you click the button, you should be referred to log.txt, which contains:

• The build target: uname -a
• The build log

Future: set up a build farm with a couple of VirtualBox appliances. Make the farm compile for each target, return binaries and the build logs, and upload them to github in a clean fashion. For example:

[gijs@ubuntu test]$ tree
.
├── lin
│   ├── build
│   │   ├── arch
│   │   │   ├── x86
│   │   │   │   └── jsh-1.2.0
│   │   │   └── x86_64
│   │   │       └── jsh-1.2.0
│   │   ├── debian
│   │   │   ├── x86
│   │   │   │   └── jsh-1.2.0
│   │   │   └── x86_64
│   │   │       └── jsh-1.2.0
│   │   ├── fedora
│   │   │   ├── x86
│   │   │   │   └── jsh-1.2.0
│   │   │   └── x86_64
│   │   │       └── jsh-1.2.0
│   │   ├── redhat
│   │   │   ├── x86
│   │   │   │   └── jsh-1.2.0
│   │   │   └── x86_64
│   │   │       └── jsh-1.2.0
│   │   └── ubuntu
│   │       ├── x86
│   │       │   └── jsh-1.2.0
│   │       └── x86_64
│   │           └── jsh-1.2.0
│   └── source
│       └── jsh-1.2.0.tar.gz
└── mac
    ├── build
    │   └── jsh-1.2.0.tar.gz
    └── source
        └── jsh-1.2.0.tar.gz

21 directories, 13 files

Or: use the OpenSuse Build Service

Joclock should be a builtin feature of jsh. Posting the command required to fire up joclock here for future reference.

xclock -norender -hd green -hl green -bg black -fg white -update 1 &

things todo:

• change line 11 in the Makefile to start with a tab instead of spaces
• function getprompt:
  • line gethostname(hostname, HOST_NAME_MAX); results in make error "‘HOST_NAME_MAX’ undeclared". The POSIX specification doesn't seem to be very clear here. Changing to "_POSIX_HOST_NAME_MAX" compiles on Mac.
  • Install the GNU readline library using homebrew (http://brew.sh/): `brew install readline --> hisotry not working
  • Apparently MacOS didn't have strnlen on 10.6 and earlier. --> own version?
  • alias troubles? tail->next != NULL ??

if this feature is stable we release jsh 1.1.0

Since we're currently drowning in cd (autocompletion) feature requests, I think it's probably best to group them together here. See also issues #16 and #35 and #50 Moreover we should sketch an approach to handle them here

A (non-exhaustive) list of cd autocompletion work, partially ordered by importance. I also added an indicator of task-difficulty-estimation:

• easy: As I sketched out in issue #16 , we should change some of readline defaults and allow a user to override them in ~/.inputrc

• challenging: autocomplete for cd should only complete directory names, no file names. This can be tricky , since readline API sucks here, I think. This feature is probably the most useful cd enhancement however. e.g.

  $ ls directory
  a_filename a_directoryname
  $ cd a<TAB>
  $ cd a_directoryname/   # detects a_filename isn't an appropriate completion
  

• easy: not really autocompletion, rather a bug, see issue #22

• moderate: a feature request to support "cd -" and maybe some more "cd history" You're free to invent some usefull syntax here, but maybe look at zsh syntax as a starting point. See issue #33

• challenging: cd autocompletion should escape spaces in directory names, see issue #35 I started on this, but readline API is very confusing here. If someone wants to look at it --> go ahead. I think we'll need some Stackoverflow help here. Probably not very difficult once API is clear. Don't hesitate to claim this one, if you feel like it ;-)

• challenging: program name autocompletion now is only a hack, using a predefined array of program names. This array of program names should be constructed dynamically, when booting the shell, by scanning the $PATH variable.

• challenging: @GijsTimmers ' ideas for smart autocompletion: correct wrong directory names, autocomplete at deeper levels, etc. To start on this one, we should have fixed the "only completing of directories" above

I think it's probably best to implement the things above and put a hold on more cd feauture requests for the moment. If you want, you can group all "future cd work" in a (single) separate issue, but probably best not in this issue. I'll also lock the other cd issues, so all discussion happens in this central issue.

Then, finally, we need developers here! The tasks above vary in difficulty, so feel free to claim one of them and open a pull request. Comment below to "claim" an item and discuss various difficulties on implementing...

cd also updates the PWD env var, but the supplied dir can also be relative (. or ..) and should then be resolved in the PWD var. Here's the code:

        case CD: ; // (hackhackhack to allow declarions inside a switch)
            char *dir;
            if (comd->length == 1)
                dir = getenv("HOME");
            else {
                CHK_ARGC("cd",1);
                dir = comd->cmd[1];
            }
            CHK_ERR(chdir(dir), "cd");
            setenv("PWD", dir, 1);   // BUG: dir can be relative and should then be resolved
            return EXIT_SUCCESS;
            break;

things todo:

• create setenv <name> <value> built_in that calls setenv(3)
• extend the jsh grammar to include $environment_variables (new level between expr and cmd or at comd level as aliases ??)
• modify the jsh parsing code to expand $environment_variables according to the grammar
• support the extern keyword
• allow syntactic sugar to write env_var="value" instead of env env_var "value"?

Who is kevinsawicki? checkout my branch. github is using my commits with other names

when executing make install a man page should be generated

when using the shell, it is not possible to update the shell to a newer version because of next error:
"cp: cannot create regular file ‘/opt/jsh’: Text file busy"

Probably trivial using the parsefile(2) function in jsh-common

after the update about -lncurses, my system (ubuntu 14.04) is unable to compile the code:
/usr/bin/ld: cannot find -lncurses

How can I solve this?

solution and install libreadline6 and libreadline6-dev

all: jsh-common alias jsh link
    @echo "<<<< All done >>>>"

jsh-common: jsh-common.c jsh-common.h
    gcc -g -c jsh-common.c -o jsh-common.o
alias: alias.c alias.h jsh-common.h
    gcc -g -c alias.c -o alias.o
jsh: jsh-parse.c jsh.c jsh-common.h
    gcc -g -c jsh.c -o jsh.o
link: jsh-common.o jsh.o alias.o
    gcc -g jsh-common.o jsh.o alias.o -o jsh -l readline

• @Sed command in Makefile gives a compile error on Mac OS X
  • SOLVED: added '$$'\n'' after **c**, see branch macos_sed_fix, commit 84412df
• compile error: inline function 'puts_verbatim' is not defined

when running shcat on a large file: shcat < documentationMitsuba.txt a segmentation fault occurs.

this is the end of the output from gdb:

13.2.5. Taking control of the logging system
Many operations in Mitsuba will print one or more log messages during their execution. By default,
they will be printed to the console, which may be undesirable. Similar to the C++ side, it is possible
to define custom Formatter and Appender classes to interpret and direct the flow of these messages.
This is also useful to keep track of the progress of rendering jobs.
Roughly, a Formatter turns detailed information about a logging event into a human-readable
string, and a Appender routes it to some destination (e.g. by appending it to a file or a log viewer in
a graphical user interface). Here is an example of how to activate such extensions:
import  mitsuba
from  mitsuba.core  import  *
class  MyFormatter(Formatter):
        def  format(self,  logLevel,  sourceClass,  sourceThread,  message,  filename,  line):

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff78198f3 in _IO_vfprintf_internal (s=, format=, ap=ap@entry=0x7fffffffda68) at vfprintf.c:1661
1661 vfprintf.c: No such file or directory.
(gdb)

What needs to be done:

• Fix HOST_NAME_MAX in getprompt()
  • FIX: use int hostlen = sysconf(_SC_HOST_NAME_MAX);
• Point to right readline library
  • FIX:
    1. Install newest readline library via Homebrew or Macports.
    2. Then, compile with -L<path_to_readline_lib> -lreadline.
• Fix a problem with linked lists in alias.c
  • FIX: memalloc(sizeof(alias)) -> memalloc(sizeof(struct alias)) in alias()

Allow a custom prompt. Things todo:

• create new branch
• new prompt "user defined prompt string" built_in shell command
• adapting the char* getprompt(int status) function in jsh.c
  • new definition: char* getprompt(int status, char *user_prompt_string)
  • substituting special symbols in user_prompt_string:
    • %u: username
    • %h: hostname
    • %d: current directory
    • %s: status of last executed command
    • %%: procent sign
• smart truncating the prompt string to a maximum value:
  • e.g. allowing the user to specify a max lenght (eg %l max_value ?) for the cur directory and truncating it to min{curdir, max_value} and if truncated -> delete the beginning of the string until the first '/'

Updated todo list:

Todo for release 1.0:

• testing + code review (pay even more attention to vulnerabilities as the code processes user input ...) on branch CustomizingPrompt
• find a better solution for MAX_DIR_LENGTH. This should be user definable, maybe in the prompt string (as in zsh) or using yet another built_in set or so? Note the MAX_PROMPT_LENGTH can remain hardcoded as we always need a (reasonable) limit for the prompt. When a user tries to set a prompt of more then 100 chars, he's probably hacking :p

Todo for future releases:

• store the user_defined_prompt_string in an environment variable instead of a global variable that is set by a built_in
• user defined coloring for the prompt
• show the username in bold red and/or replace prompt-ending with # when sudo is activated
• themes
• more options: e.g. the zsh options
• the pwd expansion code should detect when the home directory substring and replace it with ~
• custom prompt things such as git support (would be really cool ;-) )
  • show current git branch
  • show whether things have changed since last commit
• the feature-to-undo-all-other-feature-requests is probably to allow the output of any shell command to be included in the prompt.
• cutting off the path until a user-defined level (e.g. lowest) directory
• update the man page to document the options for the prompt command

Take, for example, this directory structure:

[gijs@therion /tmp]$ tree
.
├── Appels
├── Bramen
├── Clementijnen
├── Druiven
└── Groenten
    ├── Bloemkool
    ├── Bonen
    ├── Broccoli
    └── Paprika

Note that the current working directory is still /tmp. Typing cd BloemTab should yield: cd Groenten/Bloemkool/.

In other words, when we cd to a directory, jsh should make a list of subdirectory's subdirectories (got it?). We should also look into the possibility of autocompleting parent directories:

[gijs@therion /tmp]$ cd Groenten/          
[gijs@therion Groenten]$ pwd
/tmp/Groenten

Now, pressing cd DruiTab should yield: cd ../Druiven

We should test whether this is feasible. We should make sure that jsh stays snappy. We could limit the number of subdirectories / levels if necessary.

I want to be able to install jsh system-wide. When it's installed, I might want to remove it. I should be able to do so with a script, because it's hard to find out which files I should remove by hand.

It would be nice if this option could be embedded as a flag in make or something like that. Maybe we can also put it in the installer.

Expected behaviour:

$ pwd
  /mnt/Data
$ cd Muziek/Franki<TAB>
$ cd Muziek/Frankie Knuckles\ -\ Beyond\ The\ Mix/
$ pwd
  /mnt/Data/Muziek/Frankie Knuckles - Beyond The Mix/

Actual behaviour:

$ pwd
  /mnt/Data
$ cd Muziek/Franki<TAB>
$ cd Muziek/Frankie Knuckles - Beyond The Mix/        ## No escaping!
  [DEBUG: You entered: 'cd Muziek/Frankie Knuckles - Beyond The Mix/']
  [DEBUG: alias: input resolved to: 'cd Muziek/Frankie Knuckles - Beyond The Mix/']
  jsh: cd: wrong number of arguments    (expected = 1)
  [DEBUG: built-in: executed 'cd']
  [DEBUG: freeing comd struct for 'cd']
  [DEBUG: parseexpr: expr evaluated with return value 1]
  [DEBUG: readcmd: freeing memory for: 'cd']
$ pwd
  /mnt/Data

Since the jsh code base is growing, we should think more about structuring. Currently the code is far from cohesive. This of course limits readability and makes adding new features harder.

Ideally, the code should be divided into small cohesive 'modules', communicating via C header interfaces. Just like the alias.c & alias.h files now. I therefore drafted out a first idea on paper:

Sorry for the crappy quality ;-) Also, don't worry about the arrows to much. Most of them are just #includes for jsh-common.h, containing some helper functions. I marked the important relationships in green. Since all functionality in a shell is related, I think the coupling will never be very satisfying, but at least this way it's more controlled.

As you can see, functionality is divided in different header files. This will make adding new features (e.g. new autocompleters, new built_ins, grammar enhancement, prompt customizing, ...) more local and thus also easier. Separating the command line interface in particular, will allow implementing a alternative simpler 'fall-back' modus for readline. Being more independent from readline can be nice, I think ;-) Especially if we want to port to Android, see issue #23 .

Drop your thoughts! However, since jsh is not a master-thesis, I started a zero-code policy till Thursday so I won't be starting on this yet... ;-)

the description of the project "A proof-of-concept UNIX shell implementation in C" is outdated. It is more than a proof-of-concept. If changed, this description in the code should be updated as well.