Update run-tests --help and install --help outputs

[Scott Baden]

It seems that there may be circumstances when you need to delete .nobs your intallation of upcxx, after doing a pull to get the latest updates. I'm still not 100% sure about when this might happen. Has anyone had troubles updating their upcxx installation?

John notes that "If there are any issues with the installation, it can be cleaned by running rm -r .nobs." and that " If you are putting some logic in a user-facing script, the safe bet is to always blow away nobs. If this is just for your own development in a hack/compile/hack/compile cycle, just leave nobs alive until you get an "Uncaught exception" error, then blast nobs and try agai

Yesterday I updated upcxx and I had to rebuild my app, as the loader complained; .nobs can't keep track of apps not in the upcxx tree.

So we do need to give users a warning, so they'll understand how to deal with updates. This won't affect users who are just running our official release, but only those on the bleeding edge who are building an amended repo.

[Former user]

For the moment, let's assume there aren't any bugs in nobs. In this fantastic utopia you should never need a rm -rf .nobs ever. nobs run test/... will always rebuild code correctly. If instead of nobs run you are using the install script, you'll still need to reinstall, since an installation is a static snapshot. But again, no rm -rf .nobs, so hopefully the install is super quick.

If something fails and rm -r .nobs fixes it, please create an issue.

Users will only need warning if they use nobs directly or the install --single installer. The regular full install wipes nobs every time and so should be immune to bugs.

[Scott Baden]

So we don't need to delete our existing upcxx install (the place where we put it) either. Is that correct?

What is the purpose of the --single option of install.

Thank you John

[Dan Bonachea]

So we don't need to delete our existing upcxx install (the place where we put it) either. Is that correct?

See issue #57

[Scott Baden]

Yes I will rm -rf until told otherwise. This needs to be documented clearly so users don't get confused.

[Former user]

I think there's been confusion about rm -rf .nobs and rm -rf <install-dir>. Hopefully, you shouldn't need rm- rf .nobs ever. And from now on (since I fixed issue #57) you wont need to rm -rf <install-dir> either. You will need to reinstall after updating upcxx though, I hope that much everyone agrees on.

[Scott Baden]

Glad to hear this. What is the function of the --single mode of installation?

[Mathias Jacquelin]

Scott, do you still have any issue with updating an old build ? Or can I mark this issue as resolved ?

Best,

Mathias Jacquelin Research Scientist Lawrence Berkeley National Laboratory [e-mail] 1-510-495-2605

[Scott Baden]

I have no issues with build, but I do not know what the --single options does, and I didn't see any documentation when I looked over the weekend.

[Mathias Jacquelin]

--single option is reserved to UPC++ developers. UPC++ application developers are not expected to use it. Therefore, it will be left undocumented, and removed from any help message or build instruction.

[Dan Bonachea]

In the 9/27 meeting, John agreed to update the --help output for our scripts to point at the appropriate documents:

run-tests --help: see docs/testing.md, also add a mention of CROSS to that document

install --help: see INSTALL.md

[Scott Baden]

What is it for? I’d like to know.

[Former user]

Fixed issue 55.

→ <<cset b3a616e4dd27>>

[Paul Hargrove]

@sbbaden I assume "what is is for" means you are asking about the (intentionally un-documented) --single option.
It is for upc++ internals developers to install just a single network+(opt/dbg) instance instead of the entire build.
This is not very usefull (and potentially confusing/dangerous) to a normal user, but it can save developers like John a lot of time when rebuilding to test a change.

[Scott Baden]

Thank you for finally clearing this up.