[PATCH 0/1] Generate application configuration option documentation
chrisj at rtems.org
Sun Jul 12 23:51:48 UTC 2020
On 8/7/20 3:35 pm, Sebastian Huber wrote:
> Hello Chris,
> thanks for the detailed response.
> On 08/07/2020 03:55, Chris Johns wrote:
>> Hi Sebastian,
>> Thank you for this patch. I am sorry but I would like to see this central repo
>> issue resolved before any generated files are added to any of the project's
>> I understand to some level the path you are taking and moving along but these
>> generated files are coming from a personal repo that has changes, processes and
>> tools that are not being reviewed. I think it would be wise for us to agree on
>> what is behind all this before we agree to what is being generated.
> I rescheduled some work to be able to provide this patch earlier. Richi Dubey
> asked a couple of questions with respect to the scheduler implementation and
> some of them are related to how application configuration options define
> internal scheduler data structures. So, It would be nice to be able to reference
> them in Doxygen. You can review this patch independent from the way it is produced.
Ah OK. Can Richi hold the changes locally until we work through the process to
bring this work into the project?
>> On 7/7/20 9:25 pm, Sebastian Huber wrote:
>>> This patch adds the application configuration option documentation generated
>>> from specification items:
>> <gulp> That is a lot of files. I am not sure I am ready to accept documentation
>> as a lot of small files. I am happy to be persuaded otherwise.
>> It is not clear to me if these YAML files generate documentation, headers, are
>> requirements or something else.
> Each application configuration option and group is provided as a separate
> specification item (YAML file). The type of the specification items is
> documented here:
Thanks. I have looked at this before and it is detailed. The documentation make
sense and it is a really good starting place.
> Application configuration options are things on their own, so they have their
> own items. This enables other items to reference application configuration
When you say `item` is this in relation to the qual tools and the `item` as
> In the manual you have now a lot more cross references for example if
> you compare it with the RTEMS 4.11 documentation. Items have their own history
> and can be reviewed individually. I see this as benefits.
Yes there are benefits and I am not questioning these benefits. Once we depend
on these generated files we also depend on what generates these so this should
be our focus. Accepting the output into the repos means accepting the tools that
generated these files.
I see the process to get the repo accepted being related to the high level
architectural and procedure things this change brings. There needs to be a
review, consideration and discussion.
>> There is nothing I can see in them that defines
>> their `role` or `roles`.
> When you look at an item
> you see attributes like
> |type: interface ||interface-type: appl-config-option ||appl-config-option-type:
> integer This defines the type of an item. I corresponds to the specification
> type refinement:
> All attributes and the value types are defined by specification item types. The
> specification item types are defined by specification items:
> https://git.rtems.org/sebh/rtems-qual.git/tree/spec/spec The type information is
> used to generate the documentation in the RTEMS Software Engineering manual and
> is used to verify that the items are in line with the type specification:
> https://git.rtems.org/sebh/rtems-qual.git/tree/specverify.py |||||
>> If they do generate files how does that happen and how
>> are those generated header files related to the C code they are an interface
> The Doxygen header file of this patch is not read by the C compiler. However, I
> would like to generate also the API header files from specification items. I am
> now able to generate <rtems.h> and all header files included by this header
> which define API elements. About 585 items describe the API defined by <rtems.h>:
> This item describes rtems_semaphore_create():
> It is worth to have a closer look at this item.
> The interface specification items can be used to generate the header files with
> Doxygen markup, Sphinx sources for the manuals, and an Interface Control
> Document (ICD, a document required by ECSS).
OK. It will require a bit of time to take in the detail and to understand the
>> How do I keep each piece in sync?
> The basic work flow would be to update an item, then call one or more scripts,
> run make or ./waf, or whatever, to generate the derived content. This will alter
> the state of Git submodules. Then you create patch sets from the changes in the
> submodules and send them for review. After acceptance you push the changes and
> update the submodules.
OK, so there is no pulling changes in the repos back, it is a one way process?
>> A compile error seems a long way from one
>> of these YAML files.
> Yes, this is a bit harder to correlate. I don't think this will be a real issue.
> It is for example easy to figure out with "grep" which item is relevant.
I think you are right, coding interfaces is not done often and I think our users
are please this is the case.
>> For example, if I was to add a YAML file to this directory I have no idea where
>> I look to find the naming, what roles are assumed, how it integrates into the
>> existing documentation, doxygen or header files? I also have no idea how I would
>> create a new API header file? What if I push a new header file to the rtems.git
>> repo with a new #include in rtems.h. Does a regenerate over write it?
> Yes, this is a real issue. We have to think about how we want to continue with
> the specification stuff in general. I see clear benefits to generate content
> from a project-wide master data set to avoid the copy and paste stuff we have
> right now.
Yes this is true. We have evolved to have a number of places that share common
data, the version is an example. A script to update an item is workable but
fragile and I think it would not be long before we start to consider a master
repo that lets us work across the various locations in a unified manner.
>> It seems to me the key is the central repo and how it functions, how easy or
>> hard it is to learn about it and how well it will work in practice. A key
>> concern is taking it from a single person repo, ie you, to a project wide repo
>> with concurrent updates ad lots of moving pieces. It is easy to ask these
>> questions and I appreciate they are not all easy to answer but I think we need
>> to try.
> My approach would be to document common tasks for the every day RTEMS maintenance:
> You also have to consider the future rate of change. The API parts, application
> configuration options, and glossary terms for example are all things which don't
> change every week.
I think we will end up using something like nuggit ..
I have not looked into this tool yet. I suspect while it is just you working in
the repo it is fine but I wonder how managing a few submodules will go once it
is opened up and there are multiple people working in the files.
> With respect to the code which generates the content. It is written in Python 3
> with the new Python guidelines in mind:
> It uses yapf for the code formatting. It uses the static analyzers flake8, mypy,
> and pylint. There are unit tests using pytest which yield 100% branch coverage.
> It is not finished work. It was not reviewed, but I think it is a good start.
I have taken a quick look over the Python code and looks great. I can see it is
all formatted to the new standard and the tests look good. Thank you for this.
Following the standards we have and creating tests makes it easier for us to
accept the code. As a result I do not think we need to review it is detail.
>>> The header file is generated by the following script and module:
>>> The module uses currently a hack to resolve references external to specification
>>> items, e.g. sections in the RTEMS Classic API Guide or URLs. I think we need
>>> specialized specification items for these external references.
>> I do not understand what this means?
> Being able to generate Doxygen markup and Sphinx sources from the same data set
> was not a small task. We talk here about several weeks of work. You can't take
> it for granted that this works at all in an area with many cross references such
> as the application configuration options. For all external references I had to
> create a table in the source code:
> This is not really elegant.
The table can be moved to an external file when ready.
>>> Sebastian Huber (1):
>>> Document application configuration options
>>> cpukit/doxygen/appl-config.h | 4133 ++++++++++++++++++++++++++++++++++
>>> 1 file changed, 4133 insertions(+)
>>> create mode 100644 cpukit/doxygen/appl-config.h
>> Is this information found in the user manual as well? I am struggling to
>> understand the relationships and what is being presented where.
> Yes, you find it here for example:
More information about the devel