-
-
Notifications
You must be signed in to change notification settings - Fork 45
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
fix: patchShebangs in nonexecutable bin files (#106) #107
base: master
Are you sure you want to change the base?
Conversation
internal.nix
Outdated
@@ -331,6 +331,9 @@ rec { | |||
|
|||
${preInstallLinkCommands} | |||
|
|||
cat package.json | jq -r '. | select(.bin != null) | .bin | values[]' | while read binTarget; do |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is the package.json
guranteed to be there?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We could also pass the package.json
as 2nd argument to cat and avoid the cat invocation.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is the
package.json
guranteed to be there?
this is called for every npm package, these are required to have a package.json
We could also pass the
package.json
as 2nd argument to jq
its prettier this way ; ) but yes, lets save some nanoseconds and call only jq
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is the package.json guranteed to be there?
package.json *should* always be there
and if its missing, we should throw a fatal error
currently, the script continues and throws later
i managed to hack together a broken node_modules with missing submodules ...
in that case i got
> [email protected] preinstall /build/node_modules
> /build/node_modules/.hooks/preinstall
jq: error: Could not open file package.json: No such file or directory
patching script interpreter paths in .
but package.json should be there
$ tar tf /nix/store/gll14kwxsz9ggfhngafafh64rv4a02fr-semver-5.7.1.tgz
package/LICENSE
package/bin/semver
package/range.bnf
package/semver.js
package/package.json
package/CHANGELOG.md
package/README.md
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I also encountered a similar problem, where jq
was complaining there was no "package.json" even though it was fine accesing it with cat
and ls
when I was working on #108. I think using cat <file> | jq ...
was mitigating the problem but then I stopped encountering it after fixing a some other part of the bash code. I still have no idea why that happened/happens.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
package.json should always be there
and if its missing, we should throw a fatal error
no. there are packages without a package.json,
for example type definitions (*.d.ts) or header files (*.h)
internal.nix
Outdated
@@ -331,6 +331,9 @@ rec { | |||
|
|||
${preInstallLinkCommands} | |||
|
|||
jq -r '. | select(.bin != null) | .bin | values[]' package.json | while read binTarget; do |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
bin
field could also have string as the value if there is a single executable1
{
"name": "my-program",
"version": "1.2.5",
"bin": "./path/to/program"
}
Currently, this jq
expression doesn't seem to handle that case. I used something like 'select(has("bin")) | .bin | if type == "string" then . else .[] end'
to handle that case. But this is the first time I wrote jq
expression, so it might be far from the optimal.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
thanks!
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I am not really into maintaining jq expression.. I've done it once or twice but generally that isn't something that i want to maintain beyond a certain level of complexity.
How much effort would it be to rewrite this in a short nodejs script (without dependencies)? The conversaion so far makes me fear about any additional special cases that we will have to add in the future.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
jq is a standard tool for json manipulation, node feels like overkill for this
The conversaion so far makes me fear about any additional special cases that we will have to add in the future.
heh, not really ... there are three cases
the bin
field ...
- does not exist
- is a string
- is an object
all of these are handled
https://docs.npmjs.com/cli/v7/configuring-npm/package-json#bin
internal.nix
Outdated
@@ -331,6 +331,9 @@ rec { | |||
|
|||
${preInstallLinkCommands} | |||
|
|||
jq -r 'select(.bin != null) | .bin | if type == "string" then . else values[] end' package.json | while read binTarget; do | |||
chmod +x "$binTarget" # patchShebangs will only patch executable files |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
maybe this should throw if binTarget is not found
in my case:
npmlock2nix.shell {
src = /tmp/package;
}
the root package.json (/tmp/package/package.json) has a bin,
but npmlock2nix will copy only the *.json files to /build, so binTarget is not found
or rather, npmlock2nix should ignore bin's for the root package (/tmp/package)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
npmlock2nix should ignore bin's for the root package (/tmp/package)
even more, npmlock2nix.node_modules should ignore install scripts for the root package,
since for the root package, only the *.json files are available,
and install scripts may require other files
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is that so? I am not entirely sure that is true. Right now we do an npm install
in the node_modules
build and that does what we need (most of the time?).
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
consider this package.json
{
"name": "root-package",
"version": "1",
"scripts": {
"postinstall": "node some/script.js"
},
"dependencies": {
"foo": "1.2.3"
}
}
npmlock2nix has only two files for the root-package: package.json and package-lock.json
these are symlinked to /build in npmlock2nix.node_modules
Lines 365 to 368 in 33eb330
postPatch = '' | |
ln -sf ${patchedLockfile (sourceHashFunc githubSourceHashMap) packageLockJson} package-lock.json | |
ln -sf ${patchedPackagefile (sourceHashFunc githubSourceHashMap) packageJson} package.json | |
''; |
so some/script.js
is NOT available in /build, and npm install
would fail
the job of npmlock2nix.node_modules is only to populate node_modules
internal.nix
Outdated
@@ -283,6 +285,22 @@ rec { | |||
else | |||
throw "sourceHashFunc: spec.type '${spec.type}' is not supported. Supported types: 'github'"; | |||
|
|||
runInstallScriptsForRootPackage = '' |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
we could move this to a bash function, similar to buildPhase
in stdenv.mkDerivation
avoid unnecessary or unwanted rebuilds
With NPM 7, hook scripts, which this PR relies on, are not supported anymore, see also #110. However with #151 it will become possible to patch the sources of arbitrary packages, which should cover mostly the same use cases as hook scripts. @milahu Could you try out #151 and see if that can be used for your use cases? |
naah, thats too general
|
i guess you're confusing ...
|
# patchShebangs will only patch executable files | ||
if [[ "$(pwd)" != "/build" ]]; then # ignore the root package. bin entries only make sense for dependencies | ||
jq -r 'select(.bin != null) | .bin | if type == "string" then . else values[] end' package.json | while read binTarget; do | ||
if [[ ! -x "$binTarget" ]]; then | ||
echo "make binary executable: $(pwd)/$binTarget" | ||
chmod +x "$binTarget" || exit 1 # on error, throw ELIFECYCLE | ||
fi | ||
done | ||
fi |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@milahu At least this part of this PR relies on hook scripts, see the node_modules
buildPhase
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
aah yes, thanks
so this is part of the node_modules/.hooks/prepare
script
which makes this PR useless for npmv7
which brings me back to #110 (comment)
→ lets implement our own npm install
instead of producing workarounds for a npm-without-hooks
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think being able to patch sources is useful on its own, and it's a relatively easy and reasonable workaround for the hook problem. Patching sources is also very well-known by Nix people already.
Implementing an entirely separate npm install
feels a bit over the top to me, requiring maintenance, the potential for bugs, having to synchronize versions, adding a lot of complexity. If such global hook scripts are really needed for use cases like ours, then I think it's better to talk to npm upstream instead, urging them to add a reasonable replacement. And whether that happens or not, patching of sources will work as a workaround in any case.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think being able to patch sources is useful on its own
yes, of course. like patchPhase
in stdenv.mkDerivation
for example, npm install puppeteer
will download a full chromium binary
to node_modules/puppeteer/.local-chromium/
in nix, we can say
{
pname = "puppeteer";
patchPhase = ''
<package.json jq '.scripts.install = ""' | sponge package.json
'';
postInstall = ''
d=$out/node_modules/puppeteer/.local-chromium/linux-970485
mkdir -p $d
ln -s ${puppeteer-chromium} $d/chrome-linux
'';
related: patching json files in place
Implementing an entirely separate
npm install
feels a bit over the top to me, requiring maintenance, the potential for bugs, having to synchronize versions, adding a lot of complexity.
a working prototype is implemented in my pnpm-install-only
the complexity is limited by
- focussing only on
npm install
. its not rocket science:- verify tar.gz files
- unpack tar.gz files
- run lifecycle scripts: must be "bottom up", so that child-dependencies are installed
- populate
node_modules/.bin
- using an external library (snyk-nodejs-lockfile-parser) to parse lockfiles. bonus: this supports different lockfile formats: npm-shrinkwrap.json package-lock.json yarn.lock. pnpm-lock.yaml is supported only via workaround, but it "should" be easy to implement. (seems everybody is a frontend dev these days ...)
- using
npm run
to run lifecycle scripts. no need to reimplement that part of npm
another bonus: this custom npm install
allows to use the nix store as a machine-level store for node packages, see my nodejs-hide-symlinks
also that project is a working prototype. its just not popular, because of ... reasons. more human reasons than technical reasons
close #106