Re: [PATCH 00/45] ArmVirtPkg, OvmfPkg: list module-internal headers in INF files

[Laszlo Ersek <lersek@redhat.com>]

On 03/11/18 09:15, Ard Biesheuvel wrote:
> Hi Laszlo,
> 
> On 11 March 2018 at 01:48, Laszlo Ersek <lersek@redhat.com> wrote:
>> Repo:   https://github.com/lersek/edk2.git
>> Branch: hdr_inf_cleanup
>>
>> In
>> <http://mid.mail-archive.com/E92EE9817A31E24EB0585FDF735412F56327F7D3@ORSMSX113.amr.corp.intel.com>,
>> Mike explained why it's a good idea to list module-internal *.h files in
>> the [Sources*] sections of the INF files:
>>
>> On 11/23/15 21:28, Kinney, Michael D wrote:
>>> There are 2 reasons to list all source files used in a module build in
>>> the [Sources] section.
>>>
>>> 1) Support incremental builds.  If a change to the .h file is made,
>>>    then the module may  not be rebuilt if the .h file is not listed in
>>>    [Sources]
>>> 2) Support of UEFI Distribution Package distribution format.  The UPT
>>>    tools that creates UDP packages uses the [Sources] section for the
>>>    inventory of files.  If a file is missing, then it will not be
>>>    included in the UDP file.
>>
>> In only two years and three-four months, I've finally come around
>> addressing (1) under ArmVirtPkg and OvmfPkg.
> 
> Thanks for doing this.
> 
> However, while I highly appreciate your thoroughness and verbosity in
> most cases, I do think you've crossed a line this time :-)
> 
> Do we *really* need 4 different patches for CsmSupportLib, each adding
> a single .h file to [Sources], with an elaborate description how it is
> being used? If it is used, it needs to be listed, and if it is not, it
> needs to be removed, that's all there is to it IMO.

The structuring of the patch series reflects my thinking process and the
work I did precisely. I didn't (couldn't) investigate multiple header
files at once / in parallel; I investigated them one by one. It's easy
to squash patches, and it's hard to split them, so I maintain that
writing up and posting these patches one by one, in v1, was the right
thing to do. Personally I find it much easier to read many trivial
patches than half as many complex / divergent ones. If you prefer that I
squash patches into one per module, I can do that (I'd wait for more
feedback first though).

Second, I disagree that it's as simple as "list it if it's used". I
didn't just want to dump the .h filenames into the INF files; I wanted
to see each time whether the use of the header file was justified in the
first place -- this is not a given if there are multiple INF files in
the same directory, or an INF file has architecture-specific Sources
sections.

For example, in patch 06/45, I removed "QemuLoader.h" from "Qemu.c", and
"Qemu.c" is only built into one of the INF files under
"OvmfPkg/AcpiPlatformDxe". (Ultimately I had to list "QemuLoader.h" in
both INF files, in patch 07/45, due to "QemuFwCfgAcpi.c", which is built
into both INFs.)

For another example, in patch 37/45, I added "VbeShim.h" to
[Sources.Ia32, Sources.X64], and not to another of the [Sources*]
sections. The same applies to patch 17/45, where "X64/VirtualMemory.h"
belongs under [Sources.X64] only.

I find this is not as easy as it looks, and I meant to be thorough. If
you don't have time to wade through the patches, I'll thank you if you
ACK just the first three (ArmVirtPkg) patches.

> Apart from that, the series looks correct to me.

Thanks!
Laszlo