You need lein for running JVM tests and/or producing uberjars. For building binaries you need GraalVM. Currently we use java11-21.1.0.
To work on Babashka itself make sure Git submodules are checked out.
$ git clone https://github.com/borkdude/babashka --recursive
To update later on:
$ git submodule update --recursive
lein repl
will get you a standard REPL/nREPL connection. To work on tests use lein with-profiles +test repl
.
Add necessary classes to babashka/impl/classes.clj
. For every addition, write
a unit test, so it's clear why it is added and removing it will break the
tests. Try to reduce the size of the binary by only adding the necessary parts
of a class in :instance-check
, :constructors
, :methods
, :fields
or
:custom
.
The reflection.json
file that is needed for GraalVM compilation is generated
as part of script/uberjar
.
Test on the JVM (for development):
script/test
Test the native version:
BABASHKA_TEST_ENV=native script/test
Babashka runs tests of libraries that are compatible with it through
script/run_lib_tests
. To add tests for a new library, do the following:
- Add an entry for the library in
deps.edn
under the:lib-tests
alias. - Create a directory for the library in
test-resources/lib_tests/
and copy its tests to there. - Add an entry in
run_all_libtests.clj
to run the added test namespaces.
See build.md.
Findings from various experiments with JDBC drivers in babashka:
- Postgres: adds 3MB to the binary. It seems the maintainers have put in effort
to make the driver compatible with Graal. The driver is part of
bb
sincev0.0.89
. - Sqlite: I feel like I'm close to a working solution, but it hangs. It adds
20MB to the binary. Since sqlite has a nice CLI we could also just shell out
to it (there's an example in the examples dir). We could also build a
babashka.sqlite
namespace around the CLI maybe similar tobabashka.curl
. See #385 for details. - HSQLDB: easy to get going with Graalvm. Adds 10 MB to the binary. It's under a feature flag right now on master. See build.md for details. Derby and H2 are known to not work with GraalVM, so far this is the "best" embedded option from a Graal perspective. Setting the -Xmx value for Docker to 4500m got it to crash. 4800m did work, but it took 17 minutes (compared to 10 minutes without this feature).
- MySQL / MariaDB: can't get those to work yet. Work in progress in issue #387.
To progress work on sqlite and mySQL, I need a working Clojure example. If you want to contribute, consider making a an example Clojure GraalVM CLI that puts something in a sqlite / mysql DB and reads something from it.
Some design decisions:
- We chose the name
bb.edn
(rather thanbabashka.edn
) for the configuration file based on this poll. The namebb
combined with.edn
is not likely to cause conflicts with other tools. - We did not choose to put the babashka configuration in
deps.edn
to keep bb config isolated (and more flexible) and also support it in projects that do not usedeps.edn
- Rather than naming the home config dir
~/.bb
we chose~/.babashka
to prevent conflicts with other global tools. We might introduce a project local~/.babashka
directory for storing caches or whatnot too.
Some of these design decisions were formed in these discussions.
- Tasks do not allow passing arguments to dependent tasks, other than by rebinding
*command-line-args*
(see discussion). - Does the list of dependencies need to be dynamic? No, see discussion (same reason as args)
- bb is resolved as file > task > bb subcommand. Shadowing future subcommand is a problem that a user can solve by renaming a task or file. (same as lein aliases). Also see Conflicts.
- It is a feature that tasks are defined as top-level vars (instead of local let-bound symbols). This plays well with the Gilardi scenario, e.g. here: https://github.com/babashka/babashka.github.io/blob/ad276625f6c41f269d19450f236cb54cab2591e1/bb.edn#L7.
- The parallel option trickles down into run calls. People who use parallel will be confused if it’s dropped magically, people who don’t use parallel won’t notice anything either way so it doesn’t matter
Keep notes here about how adding libraries and classes to Babashka affects the binary size. We're registering the size of the macOS binary (as built on CircleCI).
2021/06/13 Upgrading from GraalvM 21.0 to 21.1 added roughly 3mb. Issue here.
2020/10/30 Without httpkit client+server: 68113436. With: 69503316 = 1390kb added.
2020/05/01 Removed next.jdbc
and postgres JDBC driver: 48304980
2020/04/23 Added next.jdbc
and postgres JDBC driver:
(- 51019836 48099780) = 2920kb added
2020/04/23 Added BigDecimal (- 48103868 47857732) = 246kb added
2020/04/18 Added clojure.data.xml 47808572 - 45923028 = 1886kb added.
2020/03/29 Added clj-yaml for parsing and generating yaml. 45196996 - 42626884 = 2570kb added.
2020/03/28 Added java.nio.file.FileSystem(s) to support extracting zip files 42562284 - 42021244 = 541kb added.
2020/03/22 Added java.io.FileReader 42025276 - 42008876 = 16kb added.
2020/03/20 Added transit write, writer, read, reader 42004796 - 41025212 = 980kb added (305kb zipped).
2020/03/19 Added java.lang.NumberFormatException, java.lang.RuntimeException, java.util.MissingResourceException and java.util.Properties to support cprop. 41025180 - 40729908 = 295kb added.
2020/02/21 Added java.time.temporal.ChronoUnit 40651596 - 40598260 = 53kb added.
2020/02/19, e43727955a2cdabd2bb0189c20dd7f9a18156fc9 Added fipp.edn/pprint 40598268 - 39744804 = 853kb added.
2020/02/09, c8fd1c7931d7842ebaec1fa8faf06d4ab58573bd Added java.lang.BigInteger and java.security.MessageDigest. 39281972 - 39072764 = 209kb added.
2020/04/02 v0.0.69 38883676
2020/01/24, 43eef7075f9dac038d8d28a5ee4e49b6affd9864: 38.3mb, 11.1mb zipped Added hierarchies (derive, isa?, etc).
2020/01/23, 485fef7df54d6701936704573468a1ec4c66d221: 37.4mb / 10.9mb zipped Added: StringBuilder, java.io.{Reader,Writer,PrinterWriter,PushbackReader}
2020/01/08, 303ca9e825d76a4a45bc4240a59139d342c13964: 36.9mb / 10.8mb zipped
Removing cheshire from bb: 36.2mb / 10.5mb zipped.