On 15:46:34 2019-05-01, Marius Vlad wrote:
On Wed, May 01, 2019 at 02:01:59PM +0200, Bruno Santos wrote:
On 12:21:20 2019-05-01, Marius Vlad wrote:
On Wed, May 01, 2019 at 12:28:37AM +0200, Bruno Santos wrote:
+class ErrorLevel(enum.Enum):
+ """
+ Supported error levels in inverse numerical order of severity. The
values
+ are chosen so that they map directly to a 'verbosity level'.
+ """
+ ERROR = 0
+ WARNING = 1
+ INFO = 2
+ DEBUG = 3
Is there a reason not to include all? (FATAL/NOTE)...
The current proposal is:
- ERROR: Something went terribly wrong, documentation is likely broken.
- WARNING: May have affected the documentation output in subtle ways.
Maybe some symbol is not fully defined and therefore won't be linked.
Documentation is still sound though.
- INFO: Tangentially related to documentation, but of no consequence to
the output. E.g. 'entering file xxx', 'found function token', etc.
- DEBUG: Don't affect documentation, but may be important in inspecting
what the code is doing for debug/development purposes. E.g. some
lesser error from an external module that we discarded.
I don't see what FATAL and NOTE would bring to the table.
If we really get a lot more errors and must discriminate further, I'd
argue for an horizontal expansion rather than vertical: (e.g.) -wclang,
etc. Adding more levels won't really help in my opinion.
I believe this all make sense, I was merely asking why we don't follow
exactly the clang types, giving the mapping.
Side-note: Diagnostics from clang will emit a fatal when files are not
found, and error if can't resolve data type. In our source code,
headers missing combined with function attributes will result in
symbols (functions) not being printed/showed and only the
comment is be present. Currently we tag this as ERROR, yet
documentation is rather sound, but your opinion here might
differ. Just want to make sure this behaviour is known.
Attachment:
signature.asc
Description: PGP signature