Contents of source
:
This is a source file
Contents of target.gup
:
#!/bin/bash
gup -u source
cat source > "$1"
echo "## Built by gup" >> "$1"
This example is a trivial build target. You can build it with:
$ gup -u target
This builds the file named target
in the current directory, (almost) as if
you had run: ./target.gup target
$ gup -u target
Does nothing.
The build script (target.gup
) registers its dependencies during its run. In
this case, the most recent run of target.gup
invoked gup -u source
. This
would cause source
to be build (if it were a gup
target itself), but also
causes gup
to remember that target
depends on source
. Since source
has not changed, gup
knows the file doesn't need to be rebuilt.
$ touch source
$ gup -u target
Rebuilds target
, since source
has been modified.
$ touch target
$ gup -u target
Also rebuilds target
, and prints a warning that "target" has
been modified outside of gup
. In order to ensure that all build
targets are correct and consistent, gup
will rebuild a file that
has been modified. For this reason it's not a good idea to make
manual edits to a build target - your edits will be lost next time the
target is built.
$ touch target.gup
$ gup -u target
This too will rebuild target
, since every target implicitly depends on its
own build script.
Contents of out.gup
:
#!bash
echo "tada!" > "$1"
This example shows a simple "shadow" build layout. When you ask gup
to build
a target, it will first search for a build script named <target>.gup
. If that
doesn't exist, it will look for "shadow" build scripts - a parallel directory
structure inside a folder named "gup".
This has a few benefits. For starters, it lets you separate build scripts from
build results. So here we can remove the entire build
directory without
worrying about losing any build scripts or source code.
Also, it means gup
can automatically create directories as needed. In this
example there's no build
directory in source control, gut gup
will create
it for us when we first build build/out
.
$ gup -u build/out
Since there's no build/out.gup
file, gup
will use gup/build/out.gup
as
the build script. This shadow "gup" directory can live at each level - see [the
README][README] for a more detailed description.
Contents of Gupfile
:
gup-build-obj:
build/*.o
Contents of gup-compile-obj
:
#!bash
src="src/$(basename "$2" .o).c"
gcc -o "$1" "src/$src"
Contents of main.gup
:
#!bash
SOURCES="build/main.o"
gup -u $SOURCES
ld -o "$1" $SOURCES
Contents of main.c
:
int main() {
println("Hello, world!");
return 0;
}
Not every target is a unique snowflake - in fact, most aren't.
So far we've seen how a <target>.gup
file is used to build a file named <target>
. But we often want to build a whole number of targets with the same build script. This is where the Gupfile
comes in. You can place a Gupfile
anywhere, and gup
will check it for targets relative to its location (including in subdirectories).
The Gupfile
syntax is straightforward. Each unindented line specifies a path to a build script, followed by a colon. Each subsequent indented line specifies a target, which may contain *
and **
wildards. gup
will use the named buildscript to build any target that matches one of these patterns.
Note that gup
will attempt to build any matching target. If you have source files which also match a pattern in a Gupfile
, you will need to add exclusions so that gup
doesn't erroneously try to build them. An exclusion is any pattern prefixed with an !
.
Here's an example of a Gupfile
builder which will be used to build any .o
or .md
file, but which will not be used to build README.md
(which is presumably hand-written, and not generated):
<build-script>:
*.o
*.md
!README.md
Contents of docs-metadata.tgz.gup
:
#!bash -eu
SOURCES="machine-id toc.md"
gup -u $SOURCES
tar czf "$1" $SOURCES
Contents of docs.md
:
# Getting started
Lorem impsum getting started.
# Installation
Lorem impsum intallation instructions.
Contents of machine-id.gup
:
#!bash -eu
gup --always
echo "$USER@$(hostname)" | tee "$1" | gup --contents
Contents of toc.md.gup
:
#!bash
gup -u docs
grep '^#' docs | tee "$1" | gup --contents
Normally, gup
will consider a file dirty if any of its dependencies have been touched (i.e. their mtime
has changed since the target was last built). But sometimes you might rebuild a file only to find that its contents are identical.
In cases like this, your build script can tell gup "don't consider this file to have changed unless the contents are actually different". This is done by passing in some contents to gup --contents
, either by piping the data directly to it, or by giving it the name of one or more files for it to read.
This is often called a "checksum" or "stamp" target - since it may need to be built very often, but won't cause its dependencies to be rebuilt unless it actually changes.
This example has two different stamp targets. toc.md
uses gup --contents
for efficiency - toc.md
will always be rebuilt when README.md
changes, however a target depending on toc
(e.g. docs-metadata.tgz
) will not be rebuilt unless there are changes to the section headings in README.md, since that is all toc.md
contains.
On the other hand, machine-id
is actually an impure target - it can't possibly declare what it depends on, since it grabs things from the environment (and gup
only knows how to depend on other files). In this case, the best we can do is to always build it (gup --always
), but only cause dependencies to be rebuilt if the result is different from last time.
Often the contents passed to gup --contents
will be the same contents that you write to your target, but it doesn't have to me.
Contents of Gupfile
:
gup-build-object:
build/*.o
gup-build-src:
src/.all
Contents of gup-build-src
:
#!python
from __future__ import print_function
import os, shutil, subprocess
from . import gup
gup.build(__file__)
dest, _ = sys.argv[1:]
checksum = subprocess.Popen(['gup','--contents'], stdin=subprocess.PIPE)
with open(dest, 'w') as dest:
for base, files, dirs in os.walk('src'):
# skip hidden dirs
for i in reversed(range(0, len(dirs))):
if dirs[i].startsWith('.'):
del dirs[i]
else:
# depend on each dir, so this target gets rebuilt
# when a file is created/removed
gup.build(os.path.join(base, dirs[i]))
for file in files:
# skip hidden files
if file.startsWith('.'):
continue
# print each file to dest:
path = os.path.join(base, file)
print(path, dest=dest)
# and include each file in checksum
print(path, dest=checksum.stdin)
with open(path) as src:
shutil.copyfileobj(src, checksum.stdin)
checksum.stdin.close()
assert checksum.wait() == 0, "Checksum task failed"
Contents of main.gup
:
#!python
from build_util import gup, src
from os import path
import sys, subprocess
dest, _ = sys.subprocess[1:]
sources = src.build_all()
def src_to_obj(src):
objname = path.splitext(src)[0] + '.o'
return path.join('build', objname)
objects = list(map(src_to_obj, sources))
subprocess.check_call(['ld', '-o', dest] + objects)
Contents of gup.py
:
import subprocess
def build(*targets):
subprocess.check_call(['gup','-u'] + list(targets))
build(__file__)
Contents of src.py
:
from . import gup
gup.build(__file__)
STAMP_FILE='src/.stamp'
def build_all():
# build the meta-target which depends on all src/files,
# and return its contents (one source file per line)
gup.build(STAMP_FILE)
with open(STAMP_FILE) as sources:
return sources.readlines()
Most of these examples use bash
as a build script language, because it's widely available and understandable for most users.
However, it's not necessarily a good idea to use bash
. Unless your build scripts are trivial, it can be hard to get bash
scripts right.
To show how non-bash gup
build scripts look, here's a simple set of build scripts written entirely in python. You can of course substitute python
for your preferred scripting language, or the language that your program is itself written in.
Contents of leaf-dir.gup
:
#!bash -eu
# this target has no dependencies - it'll only be rebuilt
# when the .gup file changes (e.g. we change the URL)
curl "http://example.com/archive.tgz" | tar xz -C "$1"
Contents of a.txt.gup
:
#!bash -eu
echo "ay" > "$1"
Contents of all.gup
:
#!bash -eu
gup -u a.txt b.txt
Contents of b.txt.gup
:
#!bash -eu
echo "bee" > "$1"
Most gup
files are targets. But sometimes you'll want to build a directory.
There are two distinct cases here:
These are directories which don't contain any buildable targets inside them, and which are created in their entirety by one build script. leaf-dir
is an example of this - its contents are an unpacked archive. It's fine to use a gup
script named after the directory to build these.
Because gup
builds targets cleanly and atomically, it will completely replace an old build result with the new one. If you had a parent-dir
target with sub-targets of parent-dir/a
and parent-dir/b
, that would actually mean that whenever you built parent-dir
it would completely replace this dir, including removing any previously-built versions of a
and b
. This is generally not what you want.
If you want to have a buildable directory with sub-targets, it's best to use a psuedo-file inside the directory, typically named all
(.all
is another common choice). This way, gup
doesn't try to replace the entire dir-target
directory since it's not itself a target. In this example, we also use a shadow gup/
directory so that we don't have to check an empty dir-target
into source control.
Contents of node_modules.gup
:
#!bash -eu
gup -u package.json
npm install
touch "$2"
Contents of package.json
:
{
"name": "test",
"dependencies": {
"tar": "*"
}
}
When integrating with external build tools, it's not always possible to instruct them to build a target in gup's temporary location ($1
). gup
will allow you to build directly into $2
(the target), provided:
$1
is not also created, and$2
is modified
In this case, the npm
command will populate the node_modules
directory, but there's no way to configure this, so the target is effectively writing to $2
directly.
If $1
is not created and $2
is not modified, gup
will think that nothng was built from this target, and it will remove the target. So when integrating with third-party tools which may not always modify the target, you should generally touch "$2"
to ensure gup
knows you've actually built something.