Обсуждение: Diff of this page with other version

On 2024-Mar-02, Marcos Pegoraro wrote:

> Obviously when a commit is done the committer has to add this span tag to
> that commit, so that part would be colored with green background only if
> page 16 is compared with previous ones.

I think your proposal is a reasonable idea and a very convenient service
for users ... but there's zero chance that committers are going to
accept the additional work and the resulting uglification of the
document source.

If it can be done by post-processing the XML and finding the
differences, to automatically insert some markup that lets the UI show
the differences as you suggest, then we can discuss ways to integrate
that.

-- 
Álvaro Herrera        Breisgau, Deutschland  —  https://www.EnterpriseDB.com/
"I must say, I am absolutely impressed with what pgsql's implementation of
VALUES allows me to do. It's kind of ridiculous how much "work" goes away in
my code.  Too bad I can't do this at work (Oracle 8/9)."       (Tom Allison)
           http://archives.postgresql.org/pgsql-general/2007-06/msg00016.php

On 2024-03-02 15:35 +0100, Marcos Pegoraro wrote:
> I think it is so common to all that it is difficult to see small diffs when
> navigating through doc pages of different versions of the same page. I'm
> not talking about completely new doc pages, like merge.sgml from version 14
> to 15, but those ones which have small diffs on that file. As an example
> consider copy.sgml

+1 for the general idea.

> Version 16 has this text
>     NULL '<replaceable class="parameter">null_string</replaceable>'
>     DEFAULT '<replaceable class="parameter">default_string</replaceable>'
>     HEADER [ <replaceable class="parameter">boolean</replaceable> | MATCH ]
>     QUOTE '<replaceable class="parameter">quote_character</replaceable>'
> 
> But version 15 has this
>     NULL '<replaceable class="parameter">null_string</replaceable>'
>     HEADER [ <replaceable class="parameter">boolean</replaceable> | MATCH ]
>     QUOTE '<replaceable class="parameter">quote_character</replaceable>'
> 
> As you can see, the DEFAULT line was added on version 16, but it is not
> easy to see what was changed on both versions.
> 
> Another example would be SQL/JSON Path Operators And Methods of func.sgml
> of version 16 and devel. There are new methods boolean(), decimal(),
> bigint(), timestamp(), timestamptz() and some others but they are not easy
> to see that they don't exist in version 16 but would be there when version
> 17 comes in.

I think that it's relatively easy to add version info to the <entry>
elements in func.sgml.  But only focusing on additions to the docs
totally misses changes to the text itself which is sometimes reworded
instead of just adding sama paragraphs.

> One can say that I have to read release notes before I upgrade a cluster
> because that page shows all important features and changes between
> versions. But sometimes this is a small info that just shows a better
> understanding of that feature or maybe we have several versions running and
> have doubts of what small feature exists in what version.
> 
> So, it would be interesting if we could visually see what was changed on
> both pages. Then, what I propose is something like you have when using a
> diff tool,  but in a single page, not side by side.
> 
> An easy way to do that would be add on all changed text a tag like
> <span class = "v16" style="background-color:green">Here goes changed or new
> text</span>
> 
> Obviously when a commit is done the committer has to add this span tag to
> that commit, so that part would be colored with green background only if
> page 16 is compared with previous ones.

The HTML is generated from the SGML docs.  So version info has to be
added there.  Or we use some automated tool to get the diff of the
rendered HTML.  W3C has a diff tool[1][2] based on GNU diffutils.  See
[3] for a diff of CREATE TABLE between v15 and v16.

> I know we have to put these tags to all files we already have but this can
> be done with some regex search tool to do this change to all files.

I think you underestimate the effort because the changes are not always
as clear (adding one line) as for copy.sgml as you've shown above.  For
example, if I check create_table.sgml between v15 and v16, I see that
the addition of STORAGE modified an existing line.  So that regex has to
match and wrap just the relevant substring.  When new storage modes are
added we end up with nested version info in order to compare vXX with
pre v16:

    <span class="v16">STORAGE { PLAIN | EXTERNAL | EXTENDED | MAIN | <span class="vXX">FOOBAR</span>  | DEFAULT
}</span>

> So, would this change on all doc pages be relevant ?
> If you agree with me than we can think if a tool to convert is needed or
> just a search replace is fine.

I recommend looking into [2] as a proof-of-concept.

[1] https://services.w3.org/htmldiff
[2] https://github.com/w3c/htmldiff-ui
[2]
https://services.w3.org/htmldiff?doc1=https%3A%2F%2Fwww.postgresql.org%2Fdocs%2F15%2Fsql-createtable.html&doc2=https%3A%2F%2Fwww.postgresql.org%2Fdocs%2F16%2Fsql-createtable.html

-- 
Erik

Erik Wienhold <ewie@ewie.name> writes:
> I think you underestimate the effort because the changes are not always
> as clear (adding one line) as for copy.sgml as you've shown above.  For
> example, if I check create_table.sgml between v15 and v16, I see that
> the addition of STORAGE modified an existing line.  So that regex has to
> match and wrap just the relevant substring.  When new storage modes are
> added we end up with nested version info in order to compare vXX with
> pre v16:

Yeah.  I think the chances of getting people to do this in the .sgml
files are precisely zero.  What might have a chance of happening is
to provide a way on the website of running htmldiff or similar tool
over two versions of a doc page on-the-fly and show the results.

            regards, tom lane

On 02/03/2024 17:33, Tom Lane wrote:

Erik Wienhold <ewie@ewie.name> writes:

I think you underestimate the effort because the changes are not always
as clear (adding one line) as for copy.sgml as you've shown above.  For
example, if I check create_table.sgml between v15 and v16, I see that
the addition of STORAGE modified an existing line.  So that regex has to
match and wrap just the relevant substring.  When new storage modes are
added we end up with nested version info in order to compare vXX with
pre v16:

Yeah.  I think the chances of getting people to do this in the .sgml
files are precisely zero.  What might have a chance of happening is
to provide a way on the website of running htmldiff or similar tool
over two versions of a doc page on-the-fly and show the results.

            regards, tom lane

I agree with Tom, I think automation of this process is the way to go here.

Best regards,
Jimmy

> On 2 Mar 2024, at 15:44, Alvaro Herrera <alvherre@alvh.no-ip.org> wrote:

> I think your proposal is a reasonable idea and a very convenient service
> for users ... but there's zero chance that committers are going to
> accept the additional work and the resulting uglification of the
> document source.
> 
> If it can be done by post-processing the XML and finding the
> differences, to automatically insert some markup that lets the UI show
> the differences as you suggest, then we can discuss ways to integrate
> that.

I agree with all of the above.

--
Daniel Gustafsson

> On 4 Mar 2024, at 14:11, Marcos Pegoraro <marcos@f10.com.br> wrote:

> I understand your point but cannot imagine a tool that does all those points I've detailed.

Googling for "diff two webpages", taking the first hit and plugging in the
CREATE TABLE page you used as an example gave me a comparison which pulled out
the STORAGE changes very clearly.  It didn't look exactly like your mockups,
but it did the job.

> And about uglification of document source, what I propose is just another internal tag
> for each version. same way you use <command>, <literal>, etc, you would use <v15>,
> <v16> and so on, just that. So I don't think it is a uglification of document source.

Adding versionspecific tags to the docs seems like quite the backpatching
hazard, which makes this a tough sell.

--
Daniel Gustafsson

> On 4 Mar 2024, at 14:11, Marcos Pegoraro <marcos@f10.com.br> wrote:

> I understand your point but cannot imagine a tool that does all those points I've detailed.

Googling for "diff two webpages", taking the first hit and plugging in the

CREATE TABLE page you used as an example gave me a comparison which pulled out

the STORAGE changes very clearly.  It didn't look exactly like your mockups,

but it did the job.

"Euler Taveira" <euler@eulerto.com> writes:
> There is also tools like pgPedia [1] that provides a command / function
> history. Unless there is a way to automate this request as Alvaro said, it
> would impose extra work by adding tags to mark that a feature adds a keyword
> and/or value. What happen if a deprecated keyword / value was removed? The
> synopsis content can be rearranged too. I have a gut feeling that some work
> should be done manually.

There's no doubt that we could get *better* results with the addition
of manual effort.  But even htmldiff as it stands today produces
*usable* results (I tried it on a couple of versions of the COPY
reference page).  Given that we've not offered something like this
at all before, I think it's silly to put in a lot of up-front effort
in advance of seeing what the demand really is.

In any case, I concur with the other committers that there is simply
no way we're going to accept manual change markup as an additional
expectation for documentation commits.  It's hard enough already.

            regards, tom lane

Marcos Pegoraro <marcos@f10.com.br> writes:
> Probably you got a page with minimal changes.
> Try json functions and compare devel with version 12 and you'll see a mess

[ shrug... ]  No diff extending across commit e894c6183 and followups
is going to produce anything terribly helpful ... or if you think that
is possible, please show how you would have marked it up.

            regards, tom lane

Marcos Pegoraro <marcos@f10.com.br> writes:
> Em seg., 4 de mar. de 2024 às 12:12, Tom Lane <tgl@sss.pgh.pa.us> escreveu:
>> Given that we've not offered something like this
>> at all before, I think it's silly to put in a lot of up-front effort
>> in advance of seeing what the demand really is.

> Well, only saying that I brought to this discussion several committers
> is proof of how beneficial this feature would be.

Nonsense.  The committers who have commented have all said that it'd
be a totally unacceptable amount of work for dubious gain.

            regards, tom lane