Zsh Mailing List Archive
Messages sorted by: Reverse Date, Date, Thread, Author

Re: [PATCH v3] Add code to Mercurial VCS backend to show topic if there is any.

On 2020-06-20 12:43, Daniel Shahaf wrote:
Manuel Jacob wrote on Sat, 20 Jun 2020 01:31 +0200:
"Topics" is an experimental concept in Mercurial that augments the
current branching concept (called "named branches").

For more information, see the not always up-to-date Mercurial Wiki page
 Functions/VCS_Info/Backends/VCS_INFO_get_data_hg | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/Functions/VCS_Info/Backends/VCS_INFO_get_data_hg b/Functions/VCS_Info/Backends/VCS_INFO_get_data_hg
index cd5ef321d..deeb331a0 100644
--- a/Functions/VCS_Info/Backends/VCS_INFO_get_data_hg
+++ b/Functions/VCS_Info/Backends/VCS_INFO_get_data_hg
@@ -5,7 +5,7 @@

 setopt localoptions extendedglob NO_shwordsplit

-local hgbase bmfile branchfile rebasefile dirstatefile mqseriesfile \
+local hgbase bmfile branchfile topicfile rebasefile dirstatefile mqseriesfile \
     curbmfile curbm \
     mqstatusfile mqguardsfile patchdir mergedir \
     r_csetid r_lrev r_branch i_bmhash i_bmname \
@@ -27,6 +27,7 @@ mergedir="${hgbase}/.hg/merge/"
 mqstatusfile="${patchdir}/status" # currently applied patches
@@ -69,6 +70,13 @@ fi
 # If we still don't know the branch it's safe to assume default
 [[ -n ${r_branch} ]] || r_branch="default"

+# Show topic if there is any (the UI for this experimental concept is not yet +# final, but for a long time the convention has been to join the branch name
+# and the topic name by a colon)
+if [[ -f ${topicfile} && -r ${topicfile} && -s ${topicfile} ]] ; then
+    r_branch=${r_branch}:$(head -n 1 ${topicfile})

I'm happy to read the file directly, despite experimentality, given the
discussion elsethread that you'll propose a corresponding unit test to
the upstream maintainers.

Using $(head) forks, and we try to avoid forks because they're expensive
on some platforms.  In this case, you could use «IFS= read -r REPLY
<$topicfile» instead.  (The construct «$(<foo)», just above the hunk
context, is special-cased to not fork.)

Is there anything special about `REPLY`? It seems to be a convention. Could I also use e.g. TOPIC?

As written, existing set-branch-format hooks would see the ${hook_com[branch]}
input argument set to strings such as "default:mytopic".  Would this
behaviour be expected, or would people rather expect ${hook_com[branch]}
to be set to strings such as "default" even when .hg/topic exists, and
to have a new ${hook_com[foo]} key whose value would be "mytopic" (or
possibly "default:mytopic")?  This would affect the API to hooks but
would not affect the final output.

If the topic extension is enabled, the meaning of places using branch names is changed. When determining how many heads a branch has (e.g. when warning that another head will be created), each "branch:topic" (or simply "branch" if there’s no topic) is considered separately. When specifying a revision, "branch:topic" refers to the head of the branch with branch name "branch" and topic name "topic", while "branch" refers to the headmost revision with branch name "branch" but no topic.

An example:


Assume that all three changesets have the "branch" field set to "develop". B has topic "feature1" and C has topic "feature2".

* "develop" refers to A
* "develop:feature1" refers to B
* "develop:feature2" refers to C

If the user updated to B, but the prompt shows "develop" (this is currently the case), it would be misleading. I don’t know all possibilities in which the hooks use `${hook_com[branch]}`, but I’d expect that including the topic would be correct in more cases than not including it.

Speaking of which, have you tested the output with the get-revision
style set?  (Run «zstyle ':vcs_info:hg:*' get-revision true»
interactively or in your zshrc.)  As written, I expect this patch would
result in output such as «default:mytopic:c04503e1dba4:0» with the style
set, compared to «default:c04503e1dba4:0» with the style set in current
master.  I don't see a problem here; just raising the point for

It overloads ":" to mean both "delimits traditional branch name and topic name" and "delimits branch name (in the general sense) and further information". I don’t see a problem. Users should be able to distinguish the different fields.

(The latter output can be customised by setting the the hgrevformat and
branchformat styles with the 'zstyle' builtin.  In the examplar output
above, "c04503e1dba4" is %r in hgrevformat; the final "0" is %h in
hgrevformat; "c04503e1dba4:0", as an atomic unit, is %r in branchformat;
and "default"/"default:mytopic" is %b in branchformat.  The whole
shebang is then %b in the formats/actionformats styles.)

Similarly, as written, the %b expando in "formats", "actionformats", and
"branchformat" will always be replaced with a string such as
"default:mytopic".  If someone wants to change the format,¹ they won't
be able to achieve that using zstyle settings alone; they'd have to
write an appropriate vcs_info hook in their zshrc.  What do you think
of, say, making the topic name available as a separate expando in
branchformat? Again, this would not affect the default output, but make
customizations easier to implement.

¹ Use-cases for changing the format include: colouring "default" and
"mytopic" differently (compare «zstyle '*' branchformat '%B%b%%b %U%r%u'»
when the get-revision style is set); changing the colon to a space for
readability or copypasteability; and perhaps even eliding the topic
name entirely(?).

For the reasons stated above, the topic name should be included. The colon should also stay because the combined "branch:topic" can be passed as a revision identifier to various places. Coloring branch and topic differently wouldn’t be "wrong", but since the "branch:topic" mostly makes sense together, personally I’d like to have it in the same appearance.

# The working dir has uncommitted-changes if the revision ends with a +
 if [[ $r_lrev[-1] == + ]] ; then



Messages sorted by: Reverse Date, Date, Thread, Author