Re: [PATCH v6 6/9] docs: add doxygen preprocessor and related files

[Luca Fancellu <luca.fancellu@xxxxxxx>]

> On 23 Jun 2021, at 23:03, Stefano Stabellini <sstabellini@xxxxxxxxxx> wrote:
> 
> On Mon, 10 May 2021, Luca Fancellu wrote:
>> Add preprocessor called by doxygen before parsing headers,
>> it will include in every header a doxygen_include.h file
>> that provides missing defines and includes that are
>> usually passed by the compiler.
>> 
>> Add doxy_input.list that is a text file containing the
>> relative path to the source code file to be parsed by
>> doxygen. The path sould be relative to the xen folder:
>> E.g. xen/include/public/grant_table.h
>> 
>> Signed-off-by: Luca Fancellu <luca.fancellu@xxxxxxx>
>> ---
>> docs/xen-doxygen/doxy-preprocessor.py | 110 ++++++++++++++++++++++++++
>> docs/xen-doxygen/doxy_input.list      |   0
>> docs/xen-doxygen/doxygen_include.h.in |  32 ++++++++
>> 3 files changed, 142 insertions(+)
>> create mode 100755 docs/xen-doxygen/doxy-preprocessor.py
>> create mode 100644 docs/xen-doxygen/doxy_input.list
>> create mode 100644 docs/xen-doxygen/doxygen_include.h.in
>> 
>> diff --git a/docs/xen-doxygen/doxy-preprocessor.py 
>> b/docs/xen-doxygen/doxy-preprocessor.py
>> new file mode 100755
>> index 0000000000..496899d8e6
>> --- /dev/null
>> +++ b/docs/xen-doxygen/doxy-preprocessor.py
>> @@ -0,0 +1,110 @@
>> +#!/usr/bin/python3
>> +#
>> +# Copyright (c) 2021, Arm Limited.
>> +#
>> +# SPDX-License-Identifier: GPL-2.0
>> +#
>> +
>> +import os, sys, re
>> +
>> +
>> +# Variables that holds the preprocessed header text
>> +output_text = ""
>> +header_file_name = ""
>> +
>> +# Variables to enumerate the anonymous structs/unions
>> +anonymous_struct_count = 0
>> +anonymous_union_count = 0
>> +
>> +
>> +def error(text):
>> +    sys.stderr.write("{}\n".format(text))
>> +    sys.exit(1)
>> +
>> +
>> +def write_to_output(text):
>> +    sys.stdout.write(text)
>> +
>> +
>> +def insert_doxygen_header(text):
>> +    # Here the strategy is to insert the #include <doxygen_include.h> in the
>> +    # first line of the header
>> +    abspath = os.path.dirname(os.path.abspath(__file__))
>> +    text += "#include \"{}/doxygen_include.h\"\n".format(abspath)
>> +
>> +    return text
>> +
>> +
>> +def enumerate_anonymous(match):
>> +    global anonymous_struct_count
>> +    global anonymous_union_count
>> +
>> +    if "struct" in match.group(1):
>> +        label = "anonymous_struct_%d" % anonymous_struct_count
>> +        anonymous_struct_count += 1
>> +    else:
>> +        label = "anonymous_union_%d" % anonymous_union_count
>> +        anonymous_union_count += 1
>> +
>> +    return match.group(1) + " " + label + " {"
>> +
>> +
>> +def manage_anonymous_structs_unions(text):
>> +    # Match anonymous unions/structs with this pattern:
>> +    # struct/union {
>> +    #     [...]
>> +    #
>> +    # and substitute it in this way:
>> +    #
>> +    # struct anonymous_struct_# {
>> +    #     [...]
>> +    # or
>> +    # union anonymous_union_# {
>> +    #     [...]
>> +    # where # is a counter starting from zero, different between structs and
>> +    # unions
>> +    #
>> +    # We don't count anonymous union/struct that are part of a typedef 
>> because
>> +    # they don't create any issue for doxygen
>> +    text = re.sub(
>> +        "(?<!typedef\s)(struct|union)\s+?\{",
>> +        enumerate_anonymous,
>> +        text,
>> +        flags=re.S
>> +    )

Hi Stefano,

> 
> My python is a bit rusty but I thought this is really clever!
> 
> One question: given that anonymous_struct_count is local per file being
> processed, it always starts at 0 for each header. I think that is
> actually better from a documentation readability point of view.
> 
> However, is it possible that Doxygen gets confused in a case where we
> can multiple "struct anonymous_struct_0", e.g. one for grant_table.h,
> one for event_channel.h, etc. ?

Yes this is a very good point, I did some experiment and it can happen, however 
so far we didn’t notice any
problem because all the anonymous union/struct were part of other data 
structure, in that case there is no
problem at all because we have 
upper_data_structure::anonymous_{struct/union}_0/1/2…

So given the fact that is difficult to have clash, I would say we can handle 
any future case separately.

Having a global numbering can solve the issue but because the script is called 
separately for each header,
Implementing it will involve some changes, there should be a file to maintain 
the number across invocation
and so on.

Let me know what do you think about that and if in your opinion we can proceed 
the way it is now.

Cheers,

Luca