addurls: Don't pass --nosave to create (and use rev-create)

[kyleam]

In gh-3252, @mih says

addurls offer a global save switch that causes any dataset modification to not be saved. This includes the initial dataset content placed in by create as well. I don't see why this would be needed, and it is off by default already.

I don't either. @yarikoptic, this is from your d3d6610 (ENH: --no-save flag for addurls (TODO: tests), 2018-03-14). Do you recall any reason why you wanted create to honor --nosave?

This PR stops propagating --nosave to create, and updates addurls to use rev-create.

[codecov]

Codecov Report

Merging #3259 into master will increase coverage by 0.04%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #3259      +/-   ##
==========================================
+ Coverage   90.97%   91.02%   +0.04%     
==========================================
  Files         263      263              
  Lines       34125    34131       +6     
==========================================
+ Hits        31046    31068      +22     
+ Misses       3079     3063      -16

Impacted Files	Coverage Δ
datalad/plugin/addurls.py	99.68% <100%> (ø)	⬆️
datalad/plugin/tests/test_addurls.py	100% <100%> (ø)	⬆️
datalad/tests/test_log.py	99% <0%> (+0.99%)	⬆️
datalad/log.py	90% <0%> (+7.14%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 0bb31b3...58189e9. Read the comment docs.

[codecov]

Codecov Report

Merging #3259 into master will increase coverage by 0.04%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #3259      +/-   ##
==========================================
+ Coverage   90.97%   91.02%   +0.04%     
==========================================
  Files         263      263              
  Lines       34125    34131       +6     
==========================================
+ Hits        31046    31068      +22     
+ Misses       3079     3063      -16

Impacted Files	Coverage Δ
datalad/plugin/addurls.py	99.68% <100%> (ø)	⬆️
datalad/plugin/tests/test_addurls.py	100% <100%> (ø)	⬆️
datalad/tests/test_log.py	99% <0%> (+0.99%)	⬆️
datalad/log.py	90% <0%> (+7.14%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 0bb31b3...58189e9. Read the comment docs.

[yarikoptic]

the PR was #2302 which says

This pull request fixes #2280. It adds a --nosave flag to addurls so that it's possible to say datalad run 'datalad addurls ...'.

So it still sounds like a legit use-case to me - to be able to datalad run our own commands, which otherwise do not create run record, but when their invocation is "loaded" enough to warrant book keeping. Somewhat relates to #3236 (commit messages) as well in that.

[mih]

I don't understand why a dataset's config file must not be commited after creating a fresh dataset in order to be able to call run.

[yarikoptic]

The main target for nosave_opt was to not save the entire result so it could be saved by run in a single commit (with run record) in the superds. This way it was possible to catch what has changed inside of it as a result of addurls was ran. We do not have "split run" recording (i.e. this was the first commit for the run command, and this was the last one).
I guess it should be ok to commit (save) those created subdatasets (i.e. during create), but without reflecting that in the super dataset (i.e. not being invoked via dataset.rev_create("subds", ...)). If subdatasets had their individual "create" commits, should be ok and contribute to the commit in the super.
As for the top dataset - it would then loose information that it was created as a part of the addurls run.

From the other side - we do not demand config being saved to be used, so there should be no penalty from no saving during create, or is there some side-effect?

[kyleam]

the PR was #2302 which says

This pull request fixes #2280. It adds a --nosave flag to addurls so that it's possible to say datalad run 'datalad addurls ...'.

Ah, thanks for digging that up.

[mih]

The side effect would be that addurls is a (if not the) command that cannot use rev_create due to its particular demands, or would need to git reset afterwards. I dont think that is desired.

Having run be able to capture command effects that alter a dataset's history seems like a worthwhile thing. But it is not a problem that is to be solved in each and every command, but in run. @bpoldrack has faced this issue before in the context of datalad-hirni.

The general aspects of this issue could move to a dedicated issue for planning.

[yarikoptic]

spoken by a true revolutionist @mih

Весь мир насилья мы разрушим
До основанья, а затем
Мы наш, мы новый мир построим, 

--- https://ru.wikipedia.org/wiki/Интернационал_(гимн)#Официальная_версия_гимна

But I agree -- there should be a more generic way to reflect information that a series of commits correspond to a single run command. Thus #3265