Merge feature/capgen into main 2021/08/12 #391

climbfuji · 2021-08-12T15:21:22Z

Update main with the latest development in feature/capgen.

Where incompatibilities exist between capgen and ccpp_prebuild, a workaround was added on the ccpp_prebuild side. Exception: in parse_checkers.py, a temporary workaround is added to internally convert unit none to 1 to match the ccpp-physics/UFS/SCM units for dimensionless variables (none) with what capgen uses (1). A discussion has been started to eliminate those inconsistencies in the actual metadata (see ESCOMP/CCPPStandardNames#18).

Update 08/18/2021: one additional routine had to be added to parse_checkers.py that I forgot to mention previously: registered_fortran_ddt_names - this routine is required so that ccpp_prebuild.py can get access to all the registered Fortran DDTs, because it utilizes capgen's metadata parser on a file-by-file basis.

Associated PRs:

#391
earth-system-radiation/rte-rrtmgp#133
NCAR/ccpp-physics#718
NOAA-EMC/fv3atm#367
ufs-community/ufs-weather-model#745

For regression testing, see ufs-community/ufs-weather-model#745.

removed some unnecessary files from the build

modified: scripts/ccpp_prebuild_config_MusicBox.py

Kinetics

climbfuji · 2021-08-17T21:29:20Z

A careful review of this PR is probably beyond my expertise right now since I'm not very familiar with capgen. To the extent that ccpp_prebuild.py calls capgen's parsers, as long as it continues to work, I don't have a problem approving the PR though. Looking at the associated fv3atm PR, it doesn't look there are a bunch of changes to understand there, perhaps with the exception of adding ccpp_types.F90 to VARIABLE_DEFINITION_FILES in the config file. I'm assuming this is now necessary due to updates in the Fortran parser or something? Also, if the order of files in that list should have DDT definitions first, why did it not error out before this PR? Is this a new requirement (I didn't think so -- I thought this is mentioned in the docs)?

Thanks for these comments. A few answers to your questions:

The order of DDTs didn't matter in the past, because we were collecting data from the capgen.py metadata parser, which didn't check if a DDT was defined when encountering a reference to it. The consistency check happened later in ccpp_prebuild.py, after all data was collected and merged. Now, capgen.py's metadata parser does check if the DDTs are known and therefore enforces the requirement of having the correct order of the variable definition files.
ccpp_prebuild.py had the ccpp_t type (in ccpp_types.F90, ccpp_types.meta) hardcoded into the scripts instead of reading it from the files. Because of 1., this data must now be read by the capgen.py metadata parser.

In summary, these PRs correct several inconsistencies i the metadata w.r.t. derived data types that ccpp_prebuild.py simply didn't care about, but capgen.py does (for good reasons). This is not only a good thing to do, but also necessary for switching to capgen.py later.

gold2718

I do not understand the changes to parse_checkers.py. The command:

git diff NCAR/feature/capgen dom/merge_feature_capgen_into_main_20210812 -- scripts/parse_tools/parse_checkers.py

shows more changes than the unit conversion from none to 1, however, these changes are not showing up in the diffs in this PR. Please explain.
Note: NCAR stands for github.com:/NCAR and dom stands for github.com/climbfuji (git command assumes that both remotes are loaded in a sandbox).

convert_metadata.py is in feature/capgen but not in this PR. Did you decide to remove it?
metavar.py does not match feature/capgen. Why?

gold2718 · 2021-08-12T22:36:39Z

.travis.yml

+language: python
+
+python:
+  - "2.7"


When do we get to drop Python 2 support?

We don't use Python 2 anymore. I think this can be upgraded safely to Python 3 in a follow-up PR.

gold2718 · 2021-08-17T23:42:19Z

scripts/metadata_table.py

+                        # end if
+                        self.__table_type = value
+                    elif key == 'dependencies':
+                        if value.lower() == "none" or value.strip() == '':


Why is a blank dependencies line an error here?

capgen's metadata parser was creating a dependencies list that contained an item [ ' ' ] - this produced an error.

That will not be happening once you merge #395 (and should not have been happening with the change you made).

gold2718 · 2021-08-17T23:43:34Z

scripts/metadata_table.py

+                            emsg += "already been declared as None"
+                            self.__pobj.add_syntax_err(emsg)
+                        else:
+                            depends = [x.strip() for x in value.split(',') if x.strip()]


This change should be made in feature/capgen first, then merged into prebuild. Doing it this way will cause issues later on.

If there is a PR to feature/capgen that I can pull in that is identical to this change - and nothing else - then I can pull it in, fix the merge conflict, assert that nothing has changed in the codebase, and don't have to rerun the tests. I don't see however why this change is causing a problem. It can be merged back from main to feature/capgen without issues.

I don't see however why this change is causing a problem.

For one thing, it took me 20 times longer to process this PR due to this change and undocumented changes to the capgen code.

gold2718 · 2021-08-17T23:47:46Z

scripts/metadata_table.py

+        # end if
+        if self.header_type == "ddt":
+            known_ddts.append(self.title)
+        # end if


There should be code here which is missing. What happened?

I am not sure, there were several merge conflicts that I had to resolve. What is missing here?

Do a diff against the feature/capgen version. After you merge #395, there should be no difference.
For this PR to be okay, there needs to be only one change between capgen code in feature/capgen and the code in this PR. Currently, that is not true.

gold2718 · 2021-08-17T23:51:07Z

scripts/metavar.py

+    >>> VariableProperty('local_name', str, check_fn_in=check_local_name).valid_value('foo(bar)')
+    'foo(bar)'
+    >>> VariableProperty('local_name', str, check_fn_in=check_local_name).valid_value('q(:,:,index_of_water_vapor_specific_humidity)')
+    'q(:,:,index_of_water_vapor_specific_humidity)'


The variables below have been removed in feature/capgen. Why are they back?

git seems to have added this code back in apparently when I did the merge

climbfuji · 2021-08-18T04:10:53Z

I do not understand the changes to parse_checkers.py. The command:
git diff NCAR/feature/capgen dom/merge_feature_capgen_into_main_20210812 -- scripts/parse_tools/parse_checkers.py
shows more changes than the unit conversion from none to 1, however, these changes are not showing up in the diffs in this PR. Please explain.
Note: NCAR stands for github.com:/NCAR and dom stands for github.com/climbfuji (git command assumes that both remotes are loaded in a sandbox).

convert_metadata.py is in feature/capgen but not in this PR. Did you decide to remove it?

metavar.py does not match feature/capgen. Why?

convert_metadata.py is no longer needed, there is no model that uses the old metadata standard
I will need to look at the diffs of metavar.py when I get back to my work laptop to see why they don't match

Fix parsing of dependencies

… into merge_feature_capgen_into_main_20210812

…ckers.py based on code review: remove old code in main, add missing code from feature/capgen

climbfuji · 2021-08-18T14:02:31Z

@gold2718 Thank you for your thorough code review. I would appreciate if you could keep these conversation at a professional level and refrain from comments like #391 (comment).

Several of the changes that you complained about were made in main (or master at that time) in agreement and with your approval. They are not the result of a sloppy merge.

Why is metavar.py different? Because feature/capgen still does not support the active attribute. In PR Update master from gsd/develop 2020/06/30 #307, the capability to process the active attribute was added to main's metavar.py, and it was review and approved. This needs to stay to keep the system working.
Similar, in metavar.py, the update to true and false attributes were made in PR Migrate to new metadata format step 1 #205, which was reviewed and approved. This needs to stay to keep the system working (ccpp_prebuild uses it).
metadata_table.py: the changes to the documentation (example metadata table, lines 61-96) were made in PRs Add dependencies to metadata #317 and Migrate to new metadata format step 1 #205, which were reviewed and approved. They are still valid and should stay.

I made the following corrections based on your code review in commit 53aa9dc:

metadata_table.py: The change in line 377 is no longer required. I think this was needed to get past a parsing issue before line 392 was fixed in my PR (and then in your PR Fix parsing of dependencies #395, which I merged into my PR to clean up the commit history).
metadata_table.py: The additional lines 736-741 are not required. They were a leftover of the original "dependencies" implementation in the main branch, which is replaced with the feature/capgen implementation as part of this PR. I removed them
metadata_table.py: I am not sure why the lines if self.module is None didn't get merged into my branch from feature/capgen, but I added them back (now lines 778-781).
parse_checkers.py: I forgot to mention in the PR description that I needed to add one additional routine to parse_checkers.py: registered_fortran_ddt_names. This routine is required so that ccpp_prebuild.py can get access to all the registered Fortran DDTs, because it utilizes capgen's metadata parser on a file-by-file basis. I updated the PR description accordingly.
parse_checkers.py: reinstantiated the missing comment line 10 "# CCPP framework imports" and removed LITERAL_INT (line 206-ish).

The fact that we are seeing more changes that slowed down your code review is not only because of the oversights listed in 4-8, but also because feature/capgen has not been updated from main (master) for years. I am unsure why this didn't happen. In all other projects I have been working, the maintainers of feature branches are usually keeping their branches up to date with the authoritative code to avoid exactly these kind of problems, especially when they are long-term developments like feature/capgen. But maybe there was a misunderstanding and we (DTC) should have created those PRs instead of you. You would still have had to review and accept them, though.

After fixing the few merge problems described above (items 4, 5, 6, 8) and updating the PR description (item 7), the current PR must be merged as-is to keep the system capgen-metadata-parser + ccpp_prebuild working. We do not have the time and I do not see why I should be making the extra efforts to make these changes separately in feature/capgen first at this point, then merge them into my PR and fix the merge conflicts. It is natural in distributed development that changes are made in the same files in different branches, but as I said above they should have been merged back to feature/capgen immediately afterwards instead of years later.

To avoid such problems in the future, I will create a follow-up PR to bring the changes/differences described above back to feature/capgen, and I strongly encourage that we do so on a regular basis going forward.

Fortunately, after making all of the abovee changes, the auto-generated code is 100% identical to before, which means we do not have to rerun the regression tests on all nine platforms (we will rerun them on just one with two different compilers).

gold2718 · 2021-08-18T15:23:39Z

Why is metavar.py different? Because feature/capgen still does not support the active attribute. In PR Update master from gsd/develop 2020/06/30 #307, the capability to process the active attribute was added to main's metavar.py, and it was review and approved. This needs to stay to keep the system working.

Then, as part of the capgen unification project, this change should be made in the feature/capgen branch. Continually updating the main branch without contributing to the feature/capgen branch is slowing down the unification effort.

gold2718 · 2021-08-18T15:26:30Z

Similar, in metavar.py, the update to true and false attributes were made in PR Migrate to new metadata format step 1 #205, which was reviewed and approved. This needs to stay to keep the system working (ccpp_prebuild uses it).

If prebuild is using a feature not used by feature/capgen, that code should be moved to the prebuild codebase.

gold2718 · 2021-08-18T15:27:12Z

metadata_table.py: the changes to the documentation (example metadata table, lines 61-96) were made in PRs Add dependencies to metadata #317 and Migrate to new metadata format step 1 #205, which were reviewed and approved. They are still valid and should stay.

Please move these changes to feature/capgen, then merge from there.

climbfuji · 2021-08-18T15:31:11Z

Similar, in metavar.py, the update to true and false attributes were made in PR Migrate to new metadata format step 1 #205, which was reviewed and approved. This needs to stay to keep the system working (ccpp_prebuild uses it).

If prebuild is using a feature not used by feature/capgen, that code should be moved to the prebuild codebase.

No. We discussed this in the ccpp-framework meeting at that time, it was decided to generalize the false/true logic and I followed your recommendations at that time to do. This is an improvement for capgen that we agreed upon, which prebuild has adopted since then. I am not going to revert or move it.

gold2718 · 2021-08-18T15:32:13Z

The fact that we are seeing more changes that slowed down your code review is not only because of the oversights listed in 4-8, but also because feature/capgen has not been updated from main (master) for years. I am unsure why this didn't happen. In all other projects I have been working, the maintainers of feature branches are usually keeping their branches up to date with the authoritative code to avoid exactly these kind of problems, especially when they are long-term developments like feature/capgen.

I agree, when are you going to do this? I argue that these changes should have been made first in feature/capgen in order to make progress towards capgen unification.

climbfuji · 2021-08-18T15:32:37Z

Why is metavar.py different? Because feature/capgen still does not support the active attribute. In PR Update master from gsd/develop 2020/06/30 #307, the capability to process the active attribute was added to main's metavar.py, and it was review and approved. This needs to stay to keep the system working.

Then, as part of the capgen unification project, this change should be made in the feature/capgen branch. Continually updating the main branch without contributing to the feature/capgen branch is slowing down the unification effort.

I will create a PR with the active attribute changes, the true/false (valid vals) changes, and the documentation changes for feature/capgen, merge it, then formally update this PR (which must result in no changes to the code), then this should be ok.

…ble.py

climbfuji · 2021-08-18T15:53:03Z

Why is metavar.py different? Because feature/capgen still does not support the active attribute. In PR Update master from gsd/develop 2020/06/30 #307, the capability to process the active attribute was added to main's metavar.py, and it was review and approved. This needs to stay to keep the system working.

Then, as part of the capgen unification project, this change should be made in the feature/capgen branch. Continually updating the main branch without contributing to the feature/capgen branch is slowing down the unification effort.

I will create a PR with the active attribute changes, the true/false (valid vals) changes, and the documentation changes for feature/capgen, merge it, then formally update this PR (which must result in no changes to the code), then this should be ok.

The PR is here: #396 - please merge and I will pull it into this PR.

climbfuji · 2021-08-18T15:57:24Z

The fact that we are seeing more changes that slowed down your code review is not only because of the oversights listed in 4-8, but also because feature/capgen has not been updated from main (master) for years. I am unsure why this didn't happen. In all other projects I have been working, the maintainers of feature branches are usually keeping their branches up to date with the authoritative code to avoid exactly these kind of problems, especially when they are long-term developments like feature/capgen.

I agree, when are you going to do this? I argue that these changes should have been made first in feature/capgen in order to make progress towards capgen unification.

Immediately after #391 is merged.

…e_metavar_doc_metadata_table Add active attribute and update true/false logic in metavar.py, update example metadata table in metadata_table.py

… into merge_feature_capgen_into_main_20210812

climbfuji · 2021-08-18T16:16:12Z

@gold2718 after merging #396 into feature/capgen and merging feature/capgen into this branch again, the files metadata_table.py and metavar.py are identical between the head of feature/capgen and this branch/PR. I am going tor re-request your review now.

gold2718

Looks good now, thanks!

cacraigucar and others added 30 commits June 6, 2018 13:28

init my_co and use in run

de57a40

removed some unnecessary files from the build

f5bfebc

Merge pull request #1 from fvitt/cac_devel

f5bf5b1

removed some unnecessary files from the build

works but verbose

9090172

cleanup and still using scm

57cfbee

one more commit

edf9c3a

add MICM as model

e8280c9

add scripts/ccpp_prebuild_config_MICM.py

d7a5e05

simple1

09d17dd

change from micm to ibox for driver code

34cd5aa

Added kinetics

697e8af

Remove ccpp_physics and replace with MICM_chemistry

313893e

Merge branch 'xx2' into cac_devel

a2bf736

change type_def filenames

8846206

remove mimc_type_def file

8bab3c4

additional script cleanup

db2f9c7

renamed ibox_type_defs file to ibox_var_defs

a35bb09

updates to support Stacy's Object-Oriented code

cbed34b

Updated to MusicBox

faf5dee

Using MusicBox_host directory

4363810

kinetics_module is now named "kinetics_module.F90"

e4cfe16

modified: scripts/ccpp_prebuild_config_MusicBox.py

Merge pull request #2 from fvitt/kinetics

5a2a20c

Kinetics

Renamed the host model variable file

ddc2a32

Adding Kessler

d4ed074

Update to ccpp-framework V2

59c0091

Merge remote-tracking branch 'origin/cleanup_ccpp-frame006'

153c38e

Update to CCPP V2

57933fc

Add k_rateConst_terminator

68497e1

Allow both 3component and terminator chemistry to be run

e40ee27

Using chemistry_driver as interface to CCPP

4fe54af

gold2718 requested changes Aug 17, 2021

View reviewed changes

Fix parsing of dependencies

6d9ca4a

climbfuji and others added 3 commits August 18, 2021 06:43

Merge pull request NCAR#395 from gold2718/capgen

eb4f5be

Fix parsing of dependencies

Merge branch 'feature/capgen' of https://github.com/NCAR/ccpp-framework…

e91b2a4

… into merge_feature_capgen_into_main_20210812

Update of scripts/metadata_table.py and scripts/parse_tools/parse_che…

53aa9dc

…ckers.py based on code review: remove old code in main, add missing code from feature/capgen

climbfuji added 3 commits August 18, 2021 09:39

Remove another legacy block of code in scripts/metavar.py

a307587

Update example metadata table in documentation in scripts/metadata_ta…

13ec708

…ble.py

Update true/false logic and add active attribute in scripts/metavar.py

eebebe6

climbfuji mentioned this pull request Aug 18, 2021

Add active attribute and update true/false logic in metavar.py, update example metadata table in metadata_table.py #396

Merged

climbfuji and others added 2 commits August 18, 2021 10:07

Merge pull request NCAR#396 from climbfuji/active_attribute_true_fals…

47ec077

…e_metavar_doc_metadata_table Add active attribute and update true/false logic in metavar.py, update example metadata table in metadata_table.py

Merge branch 'feature/capgen' of https://github.com/NCAR/ccpp-framework…

0e09eef

… into merge_feature_capgen_into_main_20210812

climbfuji requested a review from gold2718 August 18, 2021 16:16

climbfuji mentioned this pull request Aug 18, 2021

Update unit of errmsg in metavar.py (and/or metadata_parser.py, ccpp-physics, fv3atm, scm) (4 hours) #397

Closed

gold2718 approved these changes Aug 18, 2021

View reviewed changes

climbfuji merged commit 922fe44 into NCAR:main Aug 18, 2021

climbfuji mentioned this pull request Aug 18, 2021

Update feature/capgen from main 2021/08.18 #398

Merged

climbfuji deleted the merge_feature_capgen_into_main_20210812 branch June 27, 2022 03:04

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Merge feature/capgen into main 2021/08/12 #391

Merge feature/capgen into main 2021/08/12 #391

climbfuji commented Aug 12, 2021 •

edited

Loading

climbfuji commented Aug 17, 2021

gold2718 left a comment

gold2718 Aug 12, 2021

climbfuji Aug 18, 2021

gold2718 Aug 17, 2021

climbfuji Aug 18, 2021

gold2718 Aug 18, 2021

gold2718 Aug 17, 2021

climbfuji Aug 18, 2021

gold2718 Aug 18, 2021

gold2718 Aug 18, 2021 •

edited

Loading

gold2718 Aug 17, 2021

climbfuji Aug 18, 2021

gold2718 Aug 18, 2021

gold2718 Aug 17, 2021

climbfuji Aug 18, 2021

climbfuji commented Aug 18, 2021

climbfuji commented Aug 18, 2021

gold2718 commented Aug 18, 2021

gold2718 commented Aug 18, 2021

gold2718 commented Aug 18, 2021

climbfuji commented Aug 18, 2021

gold2718 commented Aug 18, 2021

climbfuji commented Aug 18, 2021

climbfuji commented Aug 18, 2021

climbfuji commented Aug 18, 2021

climbfuji commented Aug 18, 2021

gold2718 left a comment

Merge feature/capgen into main 2021/08/12 #391

Merge feature/capgen into main 2021/08/12 #391

Conversation

climbfuji commented Aug 12, 2021 • edited Loading

climbfuji commented Aug 17, 2021

gold2718 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gold2718 Aug 18, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

climbfuji commented Aug 18, 2021

climbfuji commented Aug 18, 2021

gold2718 commented Aug 18, 2021

gold2718 commented Aug 18, 2021

gold2718 commented Aug 18, 2021

climbfuji commented Aug 18, 2021

gold2718 commented Aug 18, 2021

climbfuji commented Aug 18, 2021

climbfuji commented Aug 18, 2021

climbfuji commented Aug 18, 2021

climbfuji commented Aug 18, 2021

gold2718 left a comment

Choose a reason for hiding this comment

climbfuji commented Aug 12, 2021 •

edited

Loading

gold2718 Aug 18, 2021 •

edited

Loading