Re: [gentoo-dev] [RFC] Enable CMAKE_WARN_UNUSED_CLI by default in cmake-utils for EAPI>=6

[Andrew Savchenko <bircoph@gentoo.org>]

[-- Attachment #1: Type: text/plain, Size: 2033 bytes --]

On Mon, 02 May 2016 18:06:44 +0200 Maciej Mrozowski wrote:
> Hello,
> 
> General advise: do not convert ebuilds inheriting cmake-utils to EAPI 6 unless 
> you know what you are doing (you are fully aware of eclass behaviour removed 
> with https://bugs.gentoo.org/show_bug.cgi?id=514384).
> 
> Background:
> 
> Pre EAPI-6 cmake-utils.eclass contained certain feature to mitigate CMake 
> variable case changes done by upstream.
> This feature was explicitly removed with 
> https://bugs.gentoo.org/show_bug.cgi?id=514384 and no alternative was 
> proposed.
> It opened new area of possible ebuild regression bugs when switching to EAPI-6 
> for ebuilds inheriting cmake-utils.eclass.
> 
> Unfortunately there is common misconception, also among developers, that it's 
> sufficient to simply replace "${cmake-utils_use_with foo)" with "-DWITH_foo=ON" 
> etc.
> This is MOST OF THE TIME not the case.
> When converting cmake-utils ebuild to EAPI>=6, one needs to consult 
> CMakeLists.txt wrt case each variable is written with since CMake is case-
> sensitive and WITH_FOO != WITH_foo != WITH_Foo.
> 
> Proposal:
> 
> CMake allows warning about unused CMake variables passed by CLI. Since this is 
> how Gentoo passes ebuild configuration options, it's proposed to enable this 
> feature.
> Unfortunately it won't fail compilation but at least it gives a chance to spot 
> case mismatch when reading build output.
> 
> Future thoughts:
> 
> For better damage control it's technically possible to extend configure phase 
> of cmake-utuls eclass to check mycmakeargs against parsed package buildsystem 
> but this might not be very reliable.

For me the real confusion was from this line:

    die "${FUNCNAME[1]} is banned in EAPI 6 and later: use -D$1${arg}=\"\$(usex $2)\" instead"

It recommends to use ${arg} without any warning about case, so when I just
copied what it recommends: -DWITH_nls="$(usex nls)", I had a nice surprise
and fun debugging.

Best regards,
Andrew Savchenko

[-- Attachment #2: Type: application/pgp-signature, Size: 819 bytes --]