Junio C Hamano <gitster@xxxxxxxxx> 于2021年5月3日周一 上午10:57写道： > > ZheNing Hu <adlternative@xxxxxxxxx> writes: > > > Note that `cat-file --batch` should be a superset of `ref-filter` and > > `--pretty`, because > > it supports all objects, while `ref-filter` only supports "commit" and > > "tag" two kinds > > of objects, and `--pretty` only supports "commit" one kind of object. > > What? A ref can point at any kind of objects, not necessarily > commits and tags. %(objectname), %(objecttype), etc. obviously > are applicable to any type of object. > Thanks for correcting. I only saw tag and commit in the output of `git for-each-ref` before, and blob and tree are not processed in `grab_values` in `ref-filter.c`, which deepens my illusion. `git update-ref` can make a ref point to a blob object or tree object, although I don't know what these refs are used for... A reference to a commit object makes sense, but if it is a blob, are there any suitable scenarios? > Another thing worth noting is that ref-filter needs to deal with > traits that are not tied to any particular object, but to the ref > itself, like who its upstream is and where it would be pushed to. > Indeed need to be considered "%(push)" "%(remote)" will change "refname". > > So it may be reasonable to provide `%(notes)` to "commit","tag" > > objects in `cat-file --batch` > > in the future, now integrate them into ref-filter firstly. > > And you can attach notes to objects of any type, not limited to > commits and tags. > > >> > So in this patch, I tried to migrate --pretty=%N to --format=%(notes). > > What do you mean by "migrate"? Are you making both available? > Well, here are some incorrect statements, I just want to express that the `--format="%(notes)"` implemented in "ref-filter" has the same semantics as "--pretty=%N". > >> > diff --git a/Documentation/git-for-each-ref.txt b/Documentation/git-for-each-ref.txt > >> > index 2ae2478de706..07f037a16e13 100644 > >> > --- a/Documentation/git-for-each-ref.txt > >> > +++ b/Documentation/git-for-each-ref.txt > >> > @@ -139,6 +139,9 @@ deltabase:: > >> > given object, if it is stored as a delta. Otherwise it > >> > expands to the null object name (all zeroes). > >> > > >> > +notes:: > >> > + The notes associated with the ref. > > "The notes associated with the object pointed at by the ref". > > A note is never associated with a ref. It can only associated with > an object. > Indeed. > If we are going to have %(notes), it may be natural to desire > grabbing notes for the given object from specified notes ref, e.g. > > git for-each-ref --format="%(notes:amlog) %(subject)" zh/pretty-date-human > > may want to give the same kind of information as > > git show -s --format='%N %s' --notes=amlog zh/pretty-date-human > > The underlying notes machinery however may not be prepared to work > with more than one notes tree at the same time, so > > git for-each-ref --format="%(notes) %(subject)" --notes=amlog zh/pretty-date-human > > might be a better syntax, as --format="%(notes:X) %(notes:Y)" may > not be something you can easily support. > Here is a "bad" idea that might support this feature: use another subprocess "git notes --refs=refs/notes/amlog" :) Thanks. -- ZheNing Hu
