On Sep 19, 2019, at 12:24 PM, Laszlo Ersek <lersek@redhat.com> wrote:

On 09/19/19 11:44, Leif Lindholm wrote:

Hi Liming,

On Thu, Sep 19, 2019 at 06:19:42AM +0000, Gao, Liming wrote:

I add my comments. 

-----Original Message-----
From: Baptiste Gerondeau [mailto:baptiste.gerondeau@linaro.org]
Sent: Thursday, September 19, 2019 12:05 AM
To: devel@edk2.groups.io
Cc: ard.biesheuvel@linaro.org; leif.lindholm@linaro.org; Kinney, Michael D
<michael.d.kinney@intel.com>; Gao, Liming <liming.gao@intel.com>; Zhang,
Shenglei <shenglei.zhang@intel.com>; Baptiste Gerondeau
<baptiste.gerondeau@linaro.org>
Subject: [PATCH 0/3] Arm builds on Visual Studio

EDIT: Resending the series since I mistakenly used the wrong email,
sorry !

We are currently making an effort to make ARM (and AARCH64 eventually)
builds using Microsoft's Visual Studio Compiler (aka MSVC/MSFT).

These 3 patches correspond to an effort to make the assembler work with
MSFT, which entails :
- Feeding MSFT the RVCT .asm files, since they share syntax
requirements.

Please separate the patch. Each patch is for each package, can't cross packages. 
If so, the package maintainer can easy review the change. 

I agree with this as a general rule, but for this (hopefully never to
be repeated) operation, it makes sense to me to keep each change in
this set as one patch.

For the simple reason that the alternative leaves several unusable
commits in sequence in the repository. There is simply no way to
bisect through this change on a per-package basis.

This is after all a horrible horrible hack that lets us keep using the
.asm files provided for one toolchain family (RVCT) in a different
toolchain family (MSFT), without having to delete and re-add, losing
history in the process.

Would you be OK with an exception for this extremely unusual
situation?

(The question was posed to Liming, but I'm going to follow up here with
my own thoughts, after getting CC'd by Liming. Thanks for that BTW.)

Let's assume the changes are split up with a fine granularity. Under
that assumption, consider two cases:

(1) ARM builds on Visual Studio are broken until the last patch is
applied, but all other toolchains continue working fine throughout the
series.

versus

(2) ARM builds are broken on Visual Studio *and* on at least one other
-- preexistent -- toolchain, until the last patch in the series is applied.


Which case reflects reality?

If it's case (1), then I prefer the fine-grained patch series structure.
Nothing regresses with existent toolchains (so bisection works with
them), we just have to advertise the last patch in the series as the one
that enables Visual Studio.

If it's case (2), then I agree with the larger (multi-package) patch, as
a rare exception.

(We can also put it like this: if it's *possible* to write this series
such that it enables (1), then we should strive for that.)

Thanks,
Laszlo