depinst.py ignore flag

[sdarwin]

How does the --ignore flag work in depinst.py?

Example:

git clone https://github.com/boostorg/boost
cd boost
git submodule update --init tools/boostdep
git submodule update --init tools/quickbook
python3 tools/boostdep/depinst/depinst.py --ignore unordered ../tools/quickbook

Notice --ignore unordered

The verbose output of depinst.py -vv shows:

-N: ['unordered']

Then, would it be expected if the submodule unordered is downloaded/cloned? It is.

[pdimov]

--ignore ignores direct dependencies, but not indirect ones. This might be why.

[sdarwin]

When building docs, quickbook is necessary. So, run this:

git submodule update --init tools/quickbook
python3 tools/boostdep/depinst/depinst.py ../tools/quickbook

That installs very many boost libraries.

What if one of those is an environment where you are testing docs and trying to build them. Such as boostorg/unordered.

depinst.py clobbers the directory, overwriting it.

It would be nice for --ignore to also skip indirect dependencies.

I have implemented a work-around, for now. rebuild/recopy the lib directory after depinst runs.

[pdimov]

depinst does git submodule update, it doesn't clobber. Normally there's nothing in libs/unordered before you run depinst (because you run depinst to perform the necessary submodule updates.)

What exactly is the scenario in which there's already something in libs/unordered, and it's not what a submodule update would put there?

[sdarwin]

What exactly is the scenario

Developing, testing, and building the documentation for a boost library... when it's not necessarily matching the exact commit which is registered in the superproject.

For example, building docs for a boost library from code that has been submitted via a pull request.

Or, doing forward development, adding more features to existing documents, in a local environment.

Many build scripts seem to expect the target files are in libs/unordered (or libs/library_name). That is where they usually go. So it's a natural location for modifying and testing the doc/ files.

[sdarwin]

Maybe, what I just wrote, is not strictly necessary... You can have a repository, separate from the superproject, not in the libs folder, and then it would not hit this problem.
But for test cases I have been working with, using the structure of the superproject organization is convenient.

[pdimov]

Sure but normally when you do

git clone https://github.com/boostorg/boost
cd boost
git submodule update --init tools/boostdep
git submodule update --init tools/quickbook

there's nothing in libs/unordered. For there to be something, you have to put it there first.

So, instead of putting something there before running

python3 tools/boostdep/depinst/depinst.py ../tools/quickbook

put it there after running it.

[sdarwin]

after running it.

Yes, right. Got it.

It's creating a strict ordered dependency though. 1-2-3. That's fine for CI in github actions, and then the container is destroyed.

Some of the scripts that I wrote though, are aimed at local development, and repeated testing. The order of events isn't so determined.

[pdimov]

If we're to add this functionality, we'll need another option for it (e.g. -P --preserve). The existing --ignore means something else - it's used if you want to ignore a direct dependency including all its indirect dependencies, not when you want to preserve a specific directory.