threads / patch / 38287

v3doc: core.ignoreStat update, and clarify the --assume-unchanged effect

Subject: [PATCH v3] doc: core.ignoreStat update, and clarify the --assume-unchanged effect

## tl;dr

5 messages between Jan 5, 2015 and Jan 10, 2015. Diffs are folded; open one to read it.

replies: 4people: 2as markdown or json

Philip Oakley· Jan 5, 2015, 22:22 UTC · lore

The assume-unchanged bit, and consequently core.ignoreStat, can be misunderstood. Be assertive about the expectation that file changes should notified to Git.

Overhaul the general wording thus:
    1. direct description of what is ignored given first.
    2. example instruction of the user manual action required.
    3. use sideways indirection for assume-unchanged and update-index
       references.
    4. add a 'normally' to give leeway for the change detection.
Signed-off-by: Philip Oakley <philipoakley@iee.org>
---

This is the corrected patch which now applies on top of next and has been checked on AsciiDoc. (was $gmane/261974/focus=262022)

I have included the previous 'after three-dashes' comment and included them in the commit message. I'm happy for it to be tweaked as appropriate.

---
 Documentation/config.txt | 21 ++++++++++++---------
 1 file changed, 12 insertions(+), 9 deletions(-)
Show changes to Documentation/config.txt +12 −9
diff --git a/Documentation/config.txt b/Documentation/config.txt
index 52eeadd..fe179d0 100644
--- a/Documentation/config.txt
+++ b/Documentation/config.txt
@@ -375,15 +375,18 @@ This is useful for excluding servers inside a firewall from
 proxy use, while defaulting to a common proxy for external domains.
 
 core.ignoreStat::
-	If true, commands which modify both the working tree and the index
-	will mark the updated paths with the "assume unchanged" bit in the
-	index. These marked files are then expected to stay unchanged in the
-	working tree. If you change them you should mark their update manually.
-	Git will normally not detect the file changes by lstat() calls.
-	This is useful on systems where those calls are very slow, such as
-	cifs/Microsoft Windows.
-	See linkgit:git-update-index[1].
-	False by default.
+	If true, Git will avoid using lstat() calls to detect if files have
+	changed. Git will set the "assume-unchanged" bit for those tracked files
+	which it has updated identically in both the index and working tree.
++
+When files are modified outside of Git, the user will need to stage
+the modified files explicitly (e.g. see 'Examples' section in
+linkgit:git-update-index[1]).
+Git will not normally detect changes to those files.
++
+This is useful on systems where lstat() calls are very slow, such as
+CIFS/Microsoft Windows.
+False by default.
 
 core.preferSymlinkRefs::
 	Instead of the default "symref" format for HEAD

-- 
1.9.5.msysgit.0
Junio C Hamano· Jan 7, 2015, 18:09 UTC · re: Philip Oakley · lore

Re: [PATCH v3] doc: core.ignoreStat update, and clarify the --assume-unchanged effect

Philip Oakley <philipoakley@iee.org> writes:
Show 19 quoted lines
> The assume-unchanged bit, and consequently core.ignoreStat, can be
> misunderstood. Be assertive about the expectation that file changes should
> notified to Git.
>
> Overhaul the general wording thus:
>     1. direct description of what is ignored given first.
>     2. example instruction of the user manual action required.
>     3. use sideways indirection for assume-unchanged and update-index
>        references.
>     4. add a 'normally' to give leeway for the change detection.
>
> Signed-off-by: Philip Oakley <philipoakley@iee.org>
> ---
>
> This is the corrected patch which now applies on top of next and has been
> checked on AsciiDoc. (was $gmane/261974/focus=262022)
>
> I have included the previous 'after three-dashes' comment and included
> them in the commit message. I'm happy for it to be tweaked as appropriate.
Thanks.
Show 11 quoted lines
> diff --git a/Documentation/config.txt b/Documentation/config.txt
> index 52eeadd..fe179d0 100644
> --- a/Documentation/config.txt
> +++ b/Documentation/config.txt
> @@ -375,15 +375,18 @@ This is useful for excluding servers inside a firewall from
>  proxy use, while defaulting to a common proxy for external domains.
>  
>  core.ignoreStat::
> +	If true, Git will avoid using lstat() calls to detect if files have
> +	changed. Git will set the "assume-unchanged" bit for those tracked files
> +	which it has updated identically in both the index and working tree.

I wonder if this is better stated in two seemingly independent sentences (like your version), or "... if files have changed by setting the assume-unchanged bit ...." to clarify where the setting of the bits to these files come into the big picture, but it is minor. Either way, I think it is a lot easier to understand than what we have in 'master'.

Show 9 quoted lines
> ++
> +When files are modified outside of Git, the user will need to stage
> +the modified files explicitly (e.g. see 'Examples' section in
> +linkgit:git-update-index[1]).
> +Git will not normally detect changes to those files.
> ++
> +This is useful on systems where lstat() calls are very slow, such as
> +CIFS/Microsoft Windows.
> +False by default.

I think you are trying to make the result more readable by using separate paragraphs for separate conceptual points, but then isn't it wrong to have "False by default" as part of stating which platforms are intended targets? I wonder if we want to have that last line as its own paragraph instead.

Thanks.
>  
>  core.preferSymlinkRefs::
>  	Instead of the default "symref" format for HEAD
Philip Oakley· Jan 9, 2015, 08:44 UTC · re: Junio C Hamano · lore

Re: [PATCH v3] doc: core.ignoreStat update, and clarify the --assume-unchanged effect

From: "Junio C Hamano" <gitster@pobox.com>
Show 49 quoted lines
> Philip Oakley <philipoakley@iee.org> writes:
>
>> The assume-unchanged bit, and consequently core.ignoreStat, can be
>> misunderstood. Be assertive about the expectation that file changes 
>> should
>> notified to Git.
>>
>> Overhaul the general wording thus:
>>     1. direct description of what is ignored given first.
>>     2. example instruction of the user manual action required.
>>     3. use sideways indirection for assume-unchanged and update-index
>>        references.
>>     4. add a 'normally' to give leeway for the change detection.
>>
>> Signed-off-by: Philip Oakley <philipoakley@iee.org>
>> ---
>>
>> This is the corrected patch which now applies on top of next and has 
>> been
>> checked on AsciiDoc. (was $gmane/261974/focus=262022)
>>
>> I have included the previous 'after three-dashes' comment and 
>> included
>> them in the commit message. I'm happy for it to be tweaked as 
>> appropriate.
>
> Thanks.
>
>> diff --git a/Documentation/config.txt b/Documentation/config.txt
>> index 52eeadd..fe179d0 100644
>> --- a/Documentation/config.txt
>> +++ b/Documentation/config.txt
>> @@ -375,15 +375,18 @@ This is useful for excluding servers inside a 
>> firewall from
>>  proxy use, while defaulting to a common proxy for external domains.
>>
>>  core.ignoreStat::
>> + If true, Git will avoid using lstat() calls to detect if files have
>> + changed. Git will set the "assume-unchanged" bit for those tracked 
>> files
>> + which it has updated identically in both the index and working 
>> tree.
>
> I wonder if this is better stated in two seemingly independent
> sentences (like your version), or "... if files have changed by
> setting the assume-unchanged bit ...." to clarify where the setting
> of the bits to these files come into the big picture, but it is
> minor.  Either way, I think it is a lot easier to understand than
> what we have in 'master'.

I had considered a number of different wordings, and wanted to keep the tricky parts separate to ease cognition.

On a separte note, this patch was a development from the problem noticed by Sérgio of the different actions of 'git commit .'and 'git commit -a' when --assume-unchanged was used. Did you have any thoughts regarding Duy's patch (05 December 2014 10:56) to his code given in $gmane/260865.

I wasn't sure if it had just been missed, or if there was some other issue?

Show 15 quoted lines
>> ++
>> +When files are modified outside of Git, the user will need to stage
>> +the modified files explicitly (e.g. see 'Examples' section in
>> +linkgit:git-update-index[1]).
>> +Git will not normally detect changes to those files.
>> ++
>> +This is useful on systems where lstat() calls are very slow, such as
>> +CIFS/Microsoft Windows.
>> +False by default.
>
> I think you are trying to make the result more readable by using
> separate paragraphs for separate conceptual points, but then isn't
> it wrong to have "False by default" as part of stating which
> platforms are intended targets?  I wonder if we want to have that
> last line as its own paragraph instead.
I was happy with it being a simple separate sentence.
Show 7 quoted lines
>
> Thanks.
>
>>
>>  core.preferSymlinkRefs::
>>  Instead of the default "symref" format for HEAD
> --
Philip
Junio C Hamano· Jan 9, 2015, 19:29 UTC · re: Philip Oakley · lore

Re: [PATCH v3] doc: core.ignoreStat update, and clarify the --assume-unchanged effect

"Philip Oakley" <philipoakley@iee.org> writes:
Show 24 quoted lines
>>> diff --git a/Documentation/config.txt b/Documentation/config.txt
>>> index 52eeadd..fe179d0 100644
>>> --- a/Documentation/config.txt
>>> +++ b/Documentation/config.txt
>>> @@ -375,15 +375,18 @@ This is useful for excluding servers inside a
>>> firewall from
>>>  proxy use, while defaulting to a common proxy for external domains.
>>>
>>>  core.ignoreStat::
>>> + If true, Git will avoid using lstat() calls to detect if files have
>>> + changed. Git will set the "assume-unchanged" bit for those
>>> tracked files
>>> + which it has updated identically in both the index and working
>>> tree.
>>
>> I wonder if this is better stated in two seemingly independent
>> sentences (like your version), or "... if files have changed by
>> setting the assume-unchanged bit ...." to clarify where the setting
>> of the bits to these files come into the big picture, but it is
>> minor.  Either way, I think it is a lot easier to understand than
>> what we have in 'master'.
>
> I had considered a number of different wordings, and wanted to keep
> the tricky parts separate to ease cognition.

Hmph, but wouldn't the result get more confusing, by stating two "tricky" things in separate sentences without giving any clue to guess how these two trickies are related? That is why I suggested to say the same two things in a way that clarifies that "avoid using lstat(2)" is the effect and "setting the assume-unchanged bit" is the underlying implementation detail to cause that effect, i.e.

	If true, Git will avoid using lstat(2) calls to detect if
	files have changed by setting "assume-unchanged" bit for
	these tracked paths that Git has updated identically in both
	the index and in the working tree.
Show 8 quoted lines
> On a separte note, this patch was a development from the problem
> noticed by Sérgio of the different actions of 'git commit .'and 'git
> commit -a' when --assume-unchanged was used. Did you have any thoughts
> regarding Duy's patch (05 December 2014 10:56) to his code given in
> $gmane/260865.
>
> I wasn't sure if it had just been missed, or if there was some other
> issue?

I thought the reason why we are discussing this documentation clean-up (specifically, clarifying that assume-unchanged is a promise the user makes not to modify the paths marked as such and is not about telling Git to ignore changes to tracked paths), was because we agreed that such a change is a wrong thing to do.

It is wrong for at least two reasons.
 - The "I promise not to modify them, so please omit lstat(2)
   assuming that I keep that promise" is a performance thing---we
   shouldn't add more code to cater to people who do not keep that
   promise.
 - Adding one more case of "Git will hide changes to tracked paths
   that you promised not to change" gives more chance to confuse
   users into an incorrect understanding of what assume-unchanged
   bit is about.  By not applying $gmane/260865, we keep one more
   way for the users to notice that the bit is *not* a mechanism
   to hide changes to tracked paths.
Show 16 quoted lines
>>> +When files are modified outside of Git, the user will need to stage
>>> +the modified files explicitly (e.g. see 'Examples' section in
>>> +linkgit:git-update-index[1]).
>>> +Git will not normally detect changes to those files.
>>> ++
>>> +This is useful on systems where lstat() calls are very slow, such as
>>> +CIFS/Microsoft Windows.
>>> +False by default.
>>
>> I think you are trying to make the result more readable by using
>> separate paragraphs for separate conceptual points, but then isn't
>> it wrong to have "False by default" as part of stating which
>> platforms are intended targets?  I wonder if we want to have that
>> last line as its own paragraph instead.
>
> I was happy with it being a simple separate sentence.

I am also _for_ a separate sentence. But when a set of three paragraphs, i.e.

	A, something about A, and things about A.
        B, something about B, and things about B.
        C, something about C, and things about C.

and you want to say something X that is not specific to A or B or C, would you add that X at the end of C's paragraph, resulting in:

	A, something about A, and things about A.
        B, something about B, and things about B.
        C, something about C, and things about C.  X that applies to
        all of A, B and C.
or would it be more clear to see:
	A, something about A, and things about A.
        B, something about B, and things about B.
        C, something about C, and things about C.
        X that applies to all of A, B and C.

was the question. I think a simple separate sentence should not be part of the same "In what situations this is useful" paragraph.

Philip Oakley· Jan 10, 2015, 18:54 UTC · re: Junio C Hamano · lore

Re: [PATCH v3] doc: core.ignoreStat update, and clarify the --assume-unchanged effect

From: "Junio C Hamano" <gitster@pobox.com>
Show 40 quoted lines
> "Philip Oakley" <philipoakley@iee.org> writes:
>
>>>> diff --git a/Documentation/config.txt b/Documentation/config.txt
>>>> index 52eeadd..fe179d0 100644
>>>> --- a/Documentation/config.txt
>>>> +++ b/Documentation/config.txt
>>>> @@ -375,15 +375,18 @@ This is useful for excluding servers inside a
>>>> firewall from
>>>>  proxy use, while defaulting to a common proxy for external
>>>> domains.
>>>>
>>>>  core.ignoreStat::
>>>> + If true, Git will avoid using lstat() calls to detect if files
>>>> have
>>>> + changed. Git will set the "assume-unchanged" bit for those
>>>> tracked files
>>>> + which it has updated identically in both the index and working
>>>> tree.
>>>
>>> I wonder if this is better stated in two seemingly independent
>>> sentences (like your version), or "... if files have changed by
>>> setting the assume-unchanged bit ...." to clarify where the setting
>>> of the bits to these files come into the big picture, but it is
>>> minor.  Either way, I think it is a lot easier to understand than
>>> what we have in 'master'.
>>
>> I had considered a number of different wordings, and wanted to keep
>> the tricky parts separate to ease cognition.
>
> Hmph, but wouldn't the result get more confusing, by stating two
> "tricky" things in separate sentences without giving any clue to
> guess how these two trickies are related?  That is why I suggested
> to say the same two things in a way that clarifies that "avoid using
> lstat(2)" is the effect and "setting the assume-unchanged bit" is
> the underlying implementation detail to cause that effect, i.e.
>
> If true, Git will avoid using lstat(2) calls to detect if
> files have changed by setting "assume-unchanged" bit for
> these tracked paths that Git has updated identically in both
> the index and in the working tree.

I actually found that awkward to understand that because of both its length (as a single sentence), and who does what when being mixed together. Surely it's a 'separation of concerns' process. The splitting the sentence halves the readability score (among other things).

Show 16 quoted lines
>
>> On a separte note, this patch was a development from the problem
>> noticed by Sérgio of the different actions of 'git commit .'and 'git
>> commit -a' when --assume-unchanged was used. Did you have any
>> thoughts
>> regarding Duy's patch (05 December 2014 10:56) to his code given in
>> $gmane/260865.
>>
>> I wasn't sure if it had just been missed, or if there was some other
>> issue?
>
> I thought the reason why we are discussing this documentation
> clean-up (specifically, clarifying that assume-unchanged is a
> promise the user makes not to modify the paths marked as such and is
> not about telling Git to ignore changes to tracked paths), was
> because we agreed that such a change is a wrong thing to do.

While that is true (the documentation gave false suggestions), the code should be trying to match our hoped for optimization so that the whole tree doesn't need to be lstat'd

Show 7 quoted lines
>
> It is wrong for at least two reasons.
>
> - The "I promise not to modify them, so please omit lstat(2)
>   assuming that I keep that promise" is a performance thing---we
>   shouldn't add more code to cater to people who do not keep that
>   promise.

We should add code that keeps our "promise" that [where reasonable] we will optimise out lstats when there are large numbers of files to be checked; Surely?

Show 7 quoted lines
>
> - Adding one more case of "Git will hide changes to tracked paths
>   that you promised not to change" gives more chance to confuse
>   users into an incorrect understanding of what assume-unchanged
>   bit is about.  By not applying $gmane/260865, we keep one more
>   way for the users to notice that the bit is *not* a mechanism
>   to hide changes to tracked paths.

Surely, correcting the documentation is that route. I don't think we should be retaining the spikes on the steering wheel as a bad driver reminder. There will be other corner cases for bad drivers to spin off at.

Ultimately the reason for inclusion would be that on large repos both commit -a and commit . should be equally fast, which would need to be said in the commit message.

As a side point, a possible follow on tidy-up patch would be to have a macro to cover the common (ce->ce_flags & (CE_VALID | CE_SKIP_WORKTREE) idiom (many places, with different constructs) in the same vein as ce_skip_worktree()

Show 51 quoted lines
>
>>>> +When files are modified outside of Git, the user will need to
>>>> stage
>>>> +the modified files explicitly (e.g. see 'Examples' section in
>>>> +linkgit:git-update-index[1]).
>>>> +Git will not normally detect changes to those files.
>>>> ++
>>>> +This is useful on systems where lstat() calls are very slow, such
>>>> as
>>>> +CIFS/Microsoft Windows.
>>>> +False by default.
>>>
>>> I think you are trying to make the result more readable by using
>>> separate paragraphs for separate conceptual points, but then isn't
>>> it wrong to have "False by default" as part of stating which
>>> platforms are intended targets?  I wonder if we want to have that
>>> last line as its own paragraph instead.
>>
>> I was happy with it being a simple separate sentence.
>
> I am also _for_ a separate sentence.  But when a set of three
> paragraphs, i.e.
>
> A, something about A, and things about A.
>
>        B, something about B, and things about B.
>
>        C, something about C, and things about C.
>
> and you want to say something X that is not specific to A or B or C,
> would you add that X at the end of C's paragraph, resulting in:
>
> A, something about A, and things about A.
>
>        B, something about B, and things about B.
>
>        C, something about C, and things about C.  X that applies to
>        all of A, B and C.
>
> or would it be more clear to see:
>
> A, something about A, and things about A.
>
>        B, something about B, and things about B.
>
>        C, something about C, and things about C.
>
>        X that applies to all of A, B and C.
>
> was the question.  I think a simple separate sentence should not be
> part of the same "In what situations this is useful" paragraph.

In this case my split was about making sure that B was a distinct item within the flow, rather than being a single ABC paragraph, where the middle item gets lost in the noise [1]. People do remember the first and last item (sentence), so the default is easy to see.

-- Philip

[1] Losing the middle item(s) in a list is a common human error, and is commonly cited in human factors studies e.g. "Human Error", Reason.J, 0521314194, cites Three Mile Island maintenance errors from poor documentation among others.

← back to recent threads