r2
Free open-source Cloudflare R2 desktop client and S3 GUI for macOS, Windows, and Linux. Manage Cloudflare R2, AWS S3, M…
Dylan's File Manager
git clone https://github.com/dylanaraps/dfm.gitdylanaraps/dfmA powerful, simple and snappy terminal file manager with minimal resource usage.
Initial Announcement: https://dylan.gr/1772192922
CONFIG_SMALL: ~90KiB, CONFIG_TINY: ~40KiB, CONFIG_TINY+ -static: ~150KiB)Required:
cat, cp, date, mkdir, printf, rm, shmakeOptional:
strip (for CONFIG_SMALL and CONFIG_TINY)clang (for CONFIG_TINY)chafa (for image view using sixel)kitty (for image view kitty)$ ./configure --prefix=/usr $ make $ make DESTDIR="" install
The configure script takes three forms of arguments.
--prefix=/usr, --helpCC=/bin/cc, CFLAGS="-O3", LDFLAGS=" "-DMACRO, -DMACRO=VALUE, -UMACROThere are three different build configurations.
-O2CONFIG_SMALL: -Os + aggressive compiler flagsCONFIG_TINY: -Oz + CONFIG_SMALL + (you must set CC=clang)-static via CFLAGS.-flto via CFLAGS.Everything contained within ./configure, Makefile.in, config.h.in,
config_cmd.h.in and config_key.h.in can be configured on the command-line
via ./configure. See ./configure --help and also refer to these files for
more information.
Bonus example:
./configure \ --prefix=/usr \ -DCONFIG_TINY=1 \ CC=clang \ CFLAGS="$CFLAGS -flto -static" \ -DDFM_NO_COLOR \ -DDFM_COL_NAV="VT_SGR(34,7)"
NOTE: If you are building for an environment without support for the XTerm
alternate screen, add -DDFM_CLEAR_EDIT to your configure flags.
dfm is configured at compile-time via its config files.
./configure: Build system, compilation and installation.config.h.in: Default settings, colors, etc.config_key.h.in: Keybindings.config_cmd.h.in: Commands.Refer to these files for more information.
The config*.in files are processed by dpp (see bin/dpp) so POSIX shell
code can be used within them. Everything defined by ./configure is also
accessible within these files as variables.
See https://github.com/dylanaraps/dpp for more information.
usage: dfm [options] [path]
options:
-H | +H toggle hidden files (-H off, +H on)
-p picker mode (print selected path to stdout and exit)
-o <opener> program to use when opening files (default: xdg-open)
-c <name> position cursor over 'name' instead of first entry
-q <query> start in search results ("*query" for substring)
-s <mode> change default sort
n name
N name reverse
e extension
s size
S size reverse
d date
D date reverse
-v <mode> change default view
n name only
s size
p permissions
t time
a all
--help show this help
--version show version
path:
directory to open (default: ".")
A few things can be set at runtime via environment variables. If unset in the
environment, default values are derived from the config.h.in file.
- DFM_COPYER (The clipboard tool to use when copying PWD or file
contents. The tool is fed the data via <stdin>)
- DFM_BOOKMARK_[0-9] (Directory bookmarks. set DFM_BOOKMARK_[0-9] and then
bind act_cd_bookmark_[0-9] to the keys of your choosing)
- DFM_OPENER (Opener script to use when opening files. This could be
xdg-open or a custom script (see the script/ directory))
- DFM_TRASH (Program to use when trashing files)
- DFM_TRASH_DIR (Path to trash directory)
- DFM_IMG_MODE (Image mode to use: 'chafa' (default), 'kitty')
- DFM_SU (Privilege escalation tool to use: 'sudo' (default))
There are two ways to exit dfm.
1) act_quit (default 'q')
2) act_quit_print_pwd (default 'Q')
Exiting with 2) will make dfm output the absolute path to the directory it was
in. This output can be passed to cd to change directory automatically on exit.
$ cd "$(dfm)"
$ var=$(dfm)
$ dfm > file
dfm is a single column file-manager with VIM like keybindings. Its basic usage
is pretty straightforward and anything non-obvious can be divined by looking
at the actions each key is bound to.
The statusline is as follows:
1 1/1 [RnHE] [1+] ~0B /path/to/current/directory/<query>
1 - Shows nest level of dfm. Only shown if > 0.
1/1 - The entry number under the cursor and the total visible entries.
[RnHE] - Indicators.
R - Shown when dfm is running as root.
n - Current sort mode: [n]ame, [N]ame reverse, [s]ize,
[S]ize reverse, [d]ate modified, [D]ate modified reverse,
[e]xtension. If the current directory is too large, in place
of sort mode, [T] is shown.
H - Shown when hidden files are enabled.
E - Shown when a command fails. This indicates that the user must
check the alternate buffer (bound to 'z' by default) to see
the error messages left by the command failure.
[1+] - Number of marked files, hidden when 0.
~0B - Approximate size of directory (shallow, excludes sub-directories).
/path/to - The current directory.
/<query> - The search query if the list was filtered.
There are five view modes: Normal, Size, Permissions, Date Modified and All.
The view mode can be cycled by pressing <Tab> by default.
All is the sum of the other view modes and gives an idea of what is shown:
-rwxr-xr-x 16m 4.0K .git/
-rwxr-xr-x 2h 4.0K bin/
-rwxr-xr-x 4d 4.0K script/
-rwxr-xr-x 32m 4.0K lib/
-rwxr-xr-x 16h 4.0K platform/
-rw-r--r-- 16m 0B .config_macro.h
-rw-r--r-- 16m 62B .gitignore
-rw-r--r-- 4d 1.0K LICENSE.md
-rw-r--r-- 16m 1.8K Makefile
-rw-r--r-- 8h 1.8K Makefile.in
-rw-r--r-- 32s 6.6K README.txt
-rw-r--r-- 16m 4.0K config.h
-rw-r--r-- 32m 4.0K config.h.in
-rw-r--r-- 32m 6.5K config_cmd.h
-rw-r--r-- 32m 6.5K config_cmd.h.in
-rw-r--r-- 16m 6.5K config_key.h
-rw-r--r-- 32m 6.5K config_key.h.in
-rwxr-xr-x 16m 3.5K configure*
-rwxr-xr-x 16m 130K dfm*
-rw-r--r-- 32m 72K dfm.c
2/20 [nH] ~268K /home/dylan/kiss/fork/dfm
There are seven sort modes: name, name reverse, size, size reverse,
date modified, date modified reverse, extension. The sort mode can be
cycled by pressing '`' (backtick) by default.
The name sort performs a natural/human sort and puts directories before files.
The area where searches and commands are inputted is a complete interactive line editor supporting all the usual actions (left/right scroll, insert, bracketed clipboard paste, backspace, delete, prev/next word, etc). The default keybindings match what is found in readline and POSIXy shells.
As of now there is no <Tab> complete or up/down arrow history cycling.
NOTE: The prompt is implemented as a gap buffer. There are two buffers, cursor left and cursor right with the cursor sitting inbetween both buffers. When it comes time to commit the input it is simply joined together. Make not of this detail as it is necessary to know it when creating your own bound commands.
Images can be viewed inside of dfm by pressing i by default. This will
display the image and wait for a keypress before returning to the directory
listing. Two backends are supported: sixel (via chafa) and kitty.
The mode can be set in config.h.in or at runtime via an environment variable.
There are two search modes: startswith (default /) and substring
(default ?). They each perform a case-sensitive and incremental as-you-type
search on the current directory's entries.
Pressing <Enter> confirms the search and the results become navigable. If
there is only one match, pressing <Enter> will open the entry in a single
press.
Files can be marked and unmarked (<spacebar> by default). There are also
shortcuts to navigate between marks, select all, clear all and to invert the
selection.
The marks can be operated on in three ways.
sh -euc "<cmd>" <marks argv>)These operations are defined as "commands" which can be typed or bound to keys. To avoid copying data, only the basenames of marks are passed to commands and the commands are exec'd in the directory containing them.
Example:
cp -f %m %d -> PWD=/path/to/mark_dir cp -f a b c /path/to/pwd
Commands are simply strings which are minimally transformed into argvs and executed. Modifiers control how the string will be transformed and executed.
:echo hello -> echo hello
:echo %f world -> foreach entry: echo <entry> world
:echo %m world -> echo <entry_1> <entry_2> ... world'.
:<waycopy -> foreach entry: (stdin) waycopy
In addition to these modifiers are the following:
%p -> Path to PWD.
$WORD -> Expand environment variable.
& -> Run in background (must be last word)..
NOTE: None of the above transformations pass through or incur the cost of
running within a shell. They are merely pointer arrays passed to exec().
NOTE: %m and %f cannot be combined and only the first occurrence of %m or
%f is evaluated. Also, %m and %f must appear on their own.
If these are too limiting, prepending a ! bypasses dfm's internal command mode
and sends it all to the shell.
:!echo "$@" -> sh -euc 'echo "$@"' <entry_1> <entry_2> ... :!echo "$1" "$2" -> sh -euc 'echo "$1" "$2"' <entry_1> <entry_2> ...
Commands can be run as root by prepending sudo or a similar tool on the
command-line. For more complex situations, pressing Z by default will use
DFM_SU (default sudo) to spawn another dfm as root. The statusline will
be a different color, show the nest level and display an R indicator to make
the escalation obvious. Pressing Z again inside of this escalated mode quits
and returns to the original dfm.
This can be configured at runtime using the environment variable DFM_SU and
at compile time via the config.h.in file.
Commands can be bound to keys. When a command is bound it can either run
straight away or open the interactive prompt with pre-filled information.
Flags can also be set to better integrate the command into dfm.
Move is defined as follows:
FM_CMD(cmd_move,
.prompt = CUT(":"), - The prompt.
.left = CUT("echo mv -f %m %d"), - Text left of cursor.
.enter = fm_cmd_run, - Callback.
.config = CMD_NOT_MARK_DIR | - Forbid running in mark directory.
CMD_MUT | - Command may mutate directory.
CMD_EXEC_MARK | - Skip interactive mode if marks.
CMD_CONFLICT, - Prompt on conflicts.
)
Chown is defined as follows:
FM_CMD(cmd_chown,
.prompt = CUT(":"),
.left = CUT("chown"),
.right = CUT(" %m"), - Text right of cursor.
.enter = fm_cmd_run,
.config = CMD_MUT,
)
This opens the interactive prompt and puts the cursor between chown and %m
so the user can add additional information.
:chown | %a
In addition to fm_cmd_run, fm_cmd_run_sh can be set to bypass dfm's
internal command mode to run the command in the shell.
See the config_key.h.in and config_cmd.h.in files for more information.
I employed many tricks in order to keep memory usage low whilst still allowing for fast operations and relatively large directory trees.
When a directory too large for dfm is entered the statusline sort indicator
is replaced with [T] to signify truncation, sorting is disabled and the
statusline colored red. Truncation occurs when memory in the name storage or
entry list is exhausted, whichever comes first. The limits are reasonable and
unlikely to be reached outside of synthetic directory trees so this isn't
really a problem.
File operations using coreutils commands work well but aren't as nice as
having fully integrated internal operations. I was working on it but it ended
up being a massive pain in the ass so I abandoned the idea. It's not enough to
use the POSIX functions as you will be left fighting TOCTOU race conditions,
control flow hell, error handling madness and other crap. A solution is to
conditionally use each OS's extension functions (ie, Linux's copy_file(),
renameat2(), O_TMPFILE, AT_EMPTY_PATH, etc) but then you end up stuck in
preprocessor #ifdef soup.
UTF8 support intentionally excludes grapheme clusters, emojis and other complicated things. Everything else should work just fine though.
dfm will do partial rendering wherever possible and also tries to do as
little display IO as it can (this is what I mean by low-bandwidth in the
feature list).
The TUI is manually implemented using VT100 escape sequences and a few
optional modern ones (bracketed paste, XTerm alt screen, synchronized
updates). Look at lib/term.h, lib/term_key.h, lib/vt.h and scan dfm.c
for VT_.* to see how it works.
NOTE: dfm works in pretty much every terminal emulator in wide use but since
it intentionally doesn't use terminfo it may not display correctly in some
environments (notably the TTY console in some BSDs). I don't think there's
anything I can do to remedy this unfortunately.
The number of marks is bounded only when it comes to materializing them. For
1000 marks dfm needs the space to construct an argv to accommodate them.
This is not all, if a cd is performed, space is also needed to store the mark
entry names as the new directory will overwrite them. Marks are stored on the
end of the directory storage growing towards its middle. In other words,
materialized marks are stored in the free space not taken up by directory
entries. This creates two scenarios.
Inside the same directory as the marks dfm can mark and operate on all of
the entries without needing any extra memory as the marks are virtual.
However, if %m is used inside the mark directory, dfm must materialize
them and the number is bounded by whatever unused memory is available. This
doesn't limit operation on files as dfm will process the marks in chunks.
%f: 900 marks -> n/a -> cmd x 900%m: 900 marks -> 300 slots -> cmd x 3Outside of the directory dfm needs space to materialize the marks so
mark that travel are bounded.
In short:
%f == boundless mark operations.%m == boundless mark operations (chunked).%f == bounded mark operations.%m == bounded mark operations.I had a lot of fun writing this. Thank you for reading.
dpp: https://github.com/dylanaraps/dppmore like this
Free open-source Cloudflare R2 desktop client and S3 GUI for macOS, Windows, and Linux. Manage Cloudflare R2, AWS S3, M…
meine 🌒 - A CLI file manager and system utility built with Textual. It combines intuitive command parsing with rich t…
MTT File Manager is a Rust-based Windows file manager with tabbed navigation, rich media previews, deep system integrat…
search projects, people, and tags