Arch Linux User Repository

@HaoZeke Please post the exact error message.

I've been having trouble with VCS related build flags, kept getting set -buildvcs=false. Finally fixed it with:

diff --git i/PKGBUILD w/PKGBUILD
index a4fbb93..163516c 100644
--- i/PKGBUILD
+++ w/PKGBUILD
@@ -2,7 +2,7 @@

 _pkgname="nchat"
 pkgname="${_pkgname}-git"
-pkgver=4.41.r6.gf0e91a37
+pkgver=4.41.r12.g8f7b24e9
 pkgrel=1
 pkgdesc="console-based chat client with support for Telegram"
 url="https://github.com/d99kris/nchat"
@@ -49,7 +49,7 @@ build() {
   )

   cmake "${_cmake_options[@]}"
-  cmake --build build
+ GOFLAGS='-buildvcs=false' cmake --build build
 }

 package() {

OK, fine.

makepkg needs to control the flags to work as intended. -O3 is not the only problematic build flag. Some projects attempt to force flags that produce broken packages.

This package is working as intended. There is no reason to make any change right now. If you don't want to use ninja, run the following before building:

sed -E -e '/\b[Nn]inja\b/d' -i PKGBUILD

Points taken, my understanding of the "avoid upstream config" guidance was that it was restricted to only trouble around -O3 -- i.e. that the guidance was specifically to set CFLAGS+=(-DCMAKE_BUILD_TYPE=None), then add whatever other flags are necessary to make the project build again. But your description of make.sh breaking the build is enough to convince me that, regardless of my understanding of the general case, in this case avoiding upstream's build methods is warranted.

(The "vanilla packages" principle I alluded to is my understanding of https://wiki.archlinux.org/title/Arch_Linux#Simplicity)

(in re readline, misread pacman -Q output and thought it wasn't pulled in by base)

Only remaining sticking point is ninja -- I see it as hardcoding a build process variant that's not strictly necessary, cf the critique of the nchat package hardcoding in a ccache makedep. What are the benefits you see in it that warrant hardcoding it in?

You are incorrect either about the meaning of the "vanilla upstream" principle or its applicability. The CMake package guidelines strongly recommend setting build type to None specifically to prevent upstream developers from changing build settings, which directly opposes the principle you describe.

Custom shell scripts, including make.sh, often deviate from guidelines, which is why this package needed to be updated recently. Fixing the behavior of custom scripts requires more effort with poorer results than using cmake directly.

readline is not used during runtime, so does not belong in depends. Even if it were, it's pulled in by base, so the package is not broken by its absence.

ninja does a few things better than make by default. I am leaving it.

What guidelines does it break? IIUC, the principle is that we should packge as close to vanilla upstream as reasonable -- modifying the build mechanism seems to be in violation of that principle to me.

Which also doesn't address the use of ninja and the absence of the readline dep?

@gesh I am not using your PKGBUILD because it breaks multiple package guidelines. Removed prepare() function because it is not needed. No changes needed for depends, makedepends, build(), and package().

The following patch should address my concerns -- tested, it builds.

diff --git a/PKGBUILD b/PKGBUILD
index 51d320d..4dbff20 100644
--- a/PKGBUILD
+++ b/PKGBUILD
@@ -1,8 +1,8 @@
 # Maintainer: Your Name <johnjahi55@gmail.com>

-_pkgname="nchat"
+_pkgname=nchat
 pkgname="${_pkgname}-git"
-pkgver=4.13.r35.g317652b6
+pkgver=4.41.r6.gf0e91a37
 pkgrel=1
 pkgdesc="console-based chat client with support for Telegram"
 url="https://github.com/d99kris/nchat"
@@ -13,6 +13,7 @@ depends=(
     'file'
     'ncurses'
     'openssl'
+    'readline'
     'sqlite'
     'zlib'
 )
@@ -22,7 +23,6 @@ makedepends=(
     'git'
     'go'
     'gperf'
-  'ninja'
 )

 provides=("$_pkgname=${pkgver%%.r*}")
@@ -37,28 +37,25 @@ pkgver() {
   git describe --long --tags --abbrev=8 | sed 's/^v//;s/\([^-]*-g\)/r\1/;s/-/./g'
 }

-prepare() {
-  cd "$_pkgsrc"
-  sed -i 's|ncursesw/ncurses.h|ncurses.h|' src/ui*
-}
-
 build() {
+    cd "${srcdir}/${_pkgsrc}"
+
     local _cmake_options=(
       -B build
       -S "$_pkgsrc"
-    -G Ninja
+      -G 'Unix Makefiles'
       -DCMAKE_BUILD_TYPE=None
       -DCMAKE_INSTALL_PREFIX='/usr'
       -DCMAKE_INSTALL_MANDIR='/usr/share/man'
       -Wno-dev
     )

-  cmake "${_cmake_options[@]}"
-  cmake --build build
+    CMAKEARGS="${_cmake_options[*]}" MAKEARGS=--no-silent bash ./make.sh build
 }

 package() {
-  DESTDIR="$pkgdir" cmake --install build
+    cd "${srcdir}/${_pkgsrc}/build"

-  install -D -m644 "$_pkgsrc/LICENSE" -t "$pkgdir/usr/share/licenses/$pkgname/"
+    DESTDIR="${pkgdir}" make install
+    install -D -m644 "../LICENSE" -t "$pkgdir/usr/share/licenses/$pkgname/"
 }

• The file dep is unnecessary -- it's part of base.

• prepare() has been unnecessary for more than two years now, since f4f7da0be00ce948a9a8de75fb21fd515f796d90 explicitly shouted it out

• Unclear where the ninja dep is from -- is this a dependency, or a suggestion? Things build fine with -G 'Unix Makefiles'

• Unclear why build() implements its own logic, rather than at most running CMAKEARGS="${_cmake_options[*]}" MAKEARGS=--no-silent bash ./make.sh build

• That package() does so is more understandable, given the invocation of sudo in make.sh install

• Missing dependency: readline