erlang-history is a tiny pair of files that can be used to patch an Erlang-OTP system to add support for history in the Erlang shell before Erlang/OTP-20.

The history supported is the one available through up/down arrows on the keyboard.

Since Erlang/OTP-20rc2, Shell history is supported out of the box (although initially disabled by default) through a port of this library to the Erlang/OTP code base. Enable the shell in these versions by setting the shell_history kernel environment variable to enabled with export ERL_AFLAGS="-kernel shell_history enabled" added to your environment variables (see Configuration Options section of this readme to see more options).

Automatically (you may need to run this command with sudo):

$ make install

Manually:

1. Find out what version of the Kernel library you're using by using application:which_applications() in the Erlang shell. The version number is the last element of the tuple.
2. Compile the files (erl -make).
3. Take the .beam files in ebin/$VSN/ and move them to $ROOT/lib/kernel-$VSN/ebin/ for the OTP release of your choice.
4. Open the kernel app file ($ROOT/lib/kernel-$VSN/ebin/kernel.app) and add group_history to the modules list.
5. Start the Erlang shell associated with this version of the Erlang/OTP kernel to gain shell history.
6. In case you want to remove the patch, just recompile $ROOT$/lib/kernel-$VSN/src/group.erl, and move the resulting .beam back into the ebin/ directory. Alternatively, make a backup beforehand. Don't forget to remove the group_history module from the kernel app file's modules list.

By default, the shell history will be enabled. To disable it, the kernel application variable hist can be set to false to disable history.

Options include:

• shell_history - enabled | disabled: enables or disables shell history. Default value is enabled
• shell_history_file_bytes - 51200..N: how many bytes the shell should remember. By default, the value is set to 512kb, and the minimal value is 50kb.
• shell_history_drop - ["some", "string", ...]: lines you do not want to be saved in the history. As an example, setting hist_drop to ["q().","init:stop().","halt()."] will avoid saving most manual ways of shutting down a shell. By default, no terms are dropped.

If you are not familiar with Erlang application variables, there are two principal ways to handle them. The first one is to pass the arguments manually to erl as follows:

erl -kernel shell_history_file_bytes 120000 -kernel shell_history_drop '["q().","init:stop()."]'

The other way is to create a configuration file, looking a bit as follows:

[{kernel,[
  {shell_history_file_bytes, 120000},
  {shell_history_drop, ["q().", "init:stop()."]}
]}].

Then start the Erlang shell by doing:

erl -config hist.config

that is, if we assume hist.config is the name of your configuration file. If you're in a unix-like system, you can then alias the 'erl' command to run whatever you need:

alias erl='erl -config hist.config'

And then use it everywhere. Last but not least, if you feel dirty, you can directly find the .app file of the kernel application (in its ebin/ directory) and write the values in there.

So far I haven't planned to push this into OTP. I have no idea how reliable the code is, wrote pretty much no tests (testing this shit is hard, although unit tests would be possible) outside of trying stuff for myself.

I also feel the whole thing is a bit too hackish and I do not believe it would be up to the OTP team's standard, but if it were to be, why not? I could commit that stuff inside OTP for sure.

This branch uses disk_log instead of DETS for OTP 18 and above. This is an attempt at fixing repeated corruption issues with DETS. In doing so, it drops the per-session storage and always stores the data in the same place for a given computer. Because disk_log does not allow to just flush bits of content on rewrite (it truncates any full file), we instead use a wrap log and try to divide the configured size into up to 10 log files so that every time we rotate a log, we lose only 10% of the data. Repair should be better than those seen in DETS and the new usage should be nicer on disk.

I've used DETS tables before this point in time, as it was (at first) easy to store stuff that way. The old requests are injected into the shell when it first starts up (and it does so for all instances of the shell on a given node. Every time a new line is typed in (as the existing shell sees it fit -- I just tied myself into the existing code), it is saved into the database.

I don't know yet! I'm hesitating between MIT and BSD, although if this were to make it into Erlang/OTP, I'd go with the Erlang Public License to fit with the rest. We'll see how it goes and what people want.

Congratulations, I guess.

No, it doesn't.

It would be nice to support backwards search with ^R, but I didn't go this far into implementing things.

Users of the shell history functions (h() and v(N)) will be disappointed to know I haven't added history support for that. Although it is possible (and I have some demo code to do so), I decided it was not worth including because of all the weird problems created. These commands save the function called as Erlang term tokens (after parsing, before evaluation) and the full results of the calls. This means that a lot of data or state could be carried over sessions while no longer making sense; things like pids, sockets, etc. Rather than dealing with that, I decided it would be simpler for people to just replay the previous functions.

If you want the demo code that handled these things, let me know.

DETS repairing of tables should be properly supported. In case of a database corrupted beyond repair, removing the DB file and starting over again will work fine.

I've tested it with all quarterly versions from R13B04 up to 19.3. It worked fine for them. After that, the codebase from this repository was moved to the official OTP code base.

To be noted is that since release 17.0, the OTP team may silently release intermediary versions of Erlang/OTP on github as patches. An example of this is Erlang 17.2, which was never announced, but is installable for people who need it. Those are not supported explicitly, but pull requests are welcome.

• Fred Hebert (ferd, MononcQc).

Thanks to Robert Virding & Felix Lange for the guidance through Erlang's IO system and the fun discussions at the 2011 EUC's hackathon. It was a pretty fun day and that's where I first prototyped this.

Thanks to Richard Jones for providing the original Makefile for this and fixing a bug when generating releases, Alexander Alexeev for making the installing procedure more general, and Radosław Szymczyszyn for the fixes to make things work with R13B04.

For other bug fixes, we have to thank @jan--f and @hypernumbers