tests: normalize multiple keys on a single dict#229
Merged
slyon merged 1 commit intocanonical:mainfrom Sep 27, 2021
Merged
Conversation
4 tasks
slyon
approved these changes
Sep 27, 2021
Contributor
There was a problem hiding this comment.
Looks good! Just a small inline suggestion about the pycheckstyle formatting that we should improve.
Also, we do have the skip_generated_yaml_validation parameter to generate() if there are tests that really do not want to play nicely with the normalizer (just FYI).
The YAML normalizer has some special cases for some given keys. However, the code block where those are dealt with used `elif` everywhere, which meant that if a single dict had multiple keys needing to be dealt with, only the first one would be fixed, and the rest would continue on, still broken. This bug is latent in the current codebase, but it showed up in a patchset I'm working on, where by the game of butterfly effects the rendered configuration would insist on having proper renderer for each netdef. This key needs normalizing, but in the test cases where the netdefs have other problematic properties, for instance the GSM config in `test_cdma_config`, those pesky renderer lines would still show up and screw up the test. V2: * Beat pycheckstyle into submission in a much more sensible fashion, thanks to @slyon
schopin-pro
force-pushed
the
test-normalize-fix
branch
from
cca227c to
36cfbcb
Compare
Contributor
Author
|
Thanks for the formatting suggestion! That parameter wouldn't have been right for this, but thanks for the tip, I'll keep it in mind :) |
Codecov Report
@@ Coverage Diff @@
## main #229 +/- ##
=======================================
Coverage 99.04% 99.04%
=======================================
Files 57 57
Lines 9524 9524
=======================================
Hits 9433 9433
Misses 91 91
Continue to review full report at Codecov.
|
Contributor
|
Thanks for cleaning it up. Let's merge this! |
3 tasks
slyon
pushed a commit
that referenced
this pull request
Nov 24, 2021
Instead of directly writing into the final network state, use a separate global parser object to store any intermediate data needed during parsing. This allows us to ensure that the client can only see valid configurations, as the parser is opaque to the client, who then must call our functions to import and validate the data into the final network state. Since the public API doesn't allow us to pass around some context, we still store a static object to use in those APIs, but all the internal functions shouldn't touch any external data. This PR depends on #229 and #227 and is only a prequel to the proper YAML parser unification work, which will start with the removal of the current parsing API, replaced by one based on explicitly allocated states, in conjunction with the #232 work. To make things easier, the latter PR has been added as a dependency for this one. It also picked up a dependency on #234 along the way :) Only the commits after the merge commit are part of the PR proper, the rest are from the dependencies. COMMITS: * lib: new functions *_clear to safely clear heap-allocated complex structs We already have the functions to actually free the objects, these helpers just wrap those to also clear up the *pointer* and nullifying it. These functions are going to be used in the parser code to clear intermediate state, notably on the error paths. Function names, signature and implementations are heavily inspired by g_datalist_clear(). V2: rename the functions to match the new scheme [prefix_]type_action * parse: clear the cur_addr_options pointer once the parsing is done Keeping a pointer around after we've transferred ownership of the data to a netdef is a recipy for disaster, as we have no way of knowing if we're still responsible for cleaning up the NetplanAddressOptions object. * lib: yaml: change the error goto label to err_path `error` is usually a GError** argument, and the new name is just as legible. * lib: write_netplan_conf_full: use an early-return This allows us to shift the code left, make it easier to read. * lib: util: netplan_delete_connection: clear the state at the beginning The function is supposed to work on an already empty state, as it parses a whole tree and validates it afterwards. We thus clear it explicitly. This is a latent bug that might become apparent should the validation be a little more stringent, such as redefinition of already existing netdefs. * tests: clear the global state in teardowns to expose inter-test order dependency Some tests were relying on the state having been cleared in a previous state, which isn't necessarily the case. * cli/utils: force a full parse cycle before extracting ids by devtype While the current approach works for now, it'll break when we separate parser state and global state, until we do the actual validation. * lib: use an explicit parser context Instead of directly writing into the final network state, use a separate global parser object to store any intermediate data needed during parsing. This allows us to ensure that the client can only see valid configurations, as the parser is opaque to the client, who then must call our functions to import *and validate* the data into the final network state. Since the public API doesn't allow us to pass around some context, we still store a static object to use in those APIs, but all the internal functions shouldn't touch any external data. V2: * Remove stray cur_filename global variable * Fix some formatting * Rename 'done' field to 'parsed_defs' * Reduced the nocov zone * libnetplan: rewrite netplan_finish_parse using explicit states The netplan_finish_parse implementation is moved to our ABI attic, abi_compat.c, and is now a simple wrapper around netplan_state_import_parser_results, which is the new function used to import into a Netplan state the results of a parsing operations. As before, the idea is to maintain a constantly valid Netplan state, so this function does validation before import. * lib: expose new functions to generate YAML from Netplan state or defs The old functions have been rewritten as wrappers around the new APIs V2: * Add some header comments to the new functions * parse-nm: new API using the separate parser state * lib: add accessors to be able to reach filename from netdef ID * lib: rewrite process_input_file and process_yaml_hierarchy using the new API scheme There are no process_input_file() replacement as it can very well be written by the client code. process_yaml_hierarchy() has been replaced by a function in utils, netplan_parser_load_yaml_hierarchy(), which leaves the error policy up to the client code rather than calling exit(). The old versions have been moved to abi_compat.c and reimplemented to keep their previous semantics. * lib: move _write_netplan_conf to the abi_compat.c file This function is only there for legacy reasons, and should be replaced by a normalized API. * lib: netplan_get_filename_by_id: move to abi_compat.c This is an old API. The ABI compat version is entirely reimplemented using public, up-to-date functions. * lib: util: reimplement netplan_delete_connection with the new APIs This new implementation does away with global state, creating its own local state instead. V2: fix formatting * netplan:utils: mark the exception path of the devtype iter as nocover * netplan/cli/utils: move the iter ABI check in the wrapper class This makes it possible to use the wrapper class in autonomy, and remove the implicit dependency between the netplan_get_ids_for_devtype tests and the _NetdefIdIterator tests. The latter would crash if run before the former, as the ctypes bindings wouldn't have been initialized. * lib: networkd: fix a stray call to an old API THis call meant that the network file was generated using the global state as reference instead of the local np_state, making things such as the VLAN feature basically useless. * generate: migrate to the new libnetplan APIs This allows us to do proper cleanup on error cases. Note that since generate was the only consumer of some of the legacy APIs, those endpoints are not hit by the testsuite anymore and are thus marked as ignored by the coverage calculations.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
The YAML normalizer has some special cases for some given keys. However,
the code block where those are dealt with used
elifeverywhere, whichmeant that if a single dict had multiple keys needing to be dealt with,
only the first one would be fixed, and the rest would continue on, still
broken.
This bug is latent in the current codebase, but it showed up in a
patchset I'm working on, where by the game of butterfly effects the
rendered configuration would insist on having proper renderer for each
netdef. This key needs normalizing, but in the test cases where the
netdefs have other problematic properties, for instance the GSM config
in
test_cdma_config, those pesky renderer lines would still show upand screw up the test.
Checklist
make checksuccessfully.make check-coverage).