1 |
Hi Michael, |
2 |
|
3 |
I think I understand your points: |
4 |
- I don't understand why systems get messed up xcode-wise either, but |
5 |
I'm sure not everyone has root access to fix it |
6 |
- My current workaround works currently, indeed, dunno about the future |
7 |
- Doing more patchwork is discouraged by llvm maintainer(s?), |
8 |
didn't like my above patch, and in fact told me to leave the llvm |
9 |
ebuilds alone because the patches are not acceptable/accepted by |
10 |
upstream, if you got ideas for this... |
11 |
- Currently we don't do much cross work indeed, it's never been an aim |
12 |
either, but until the day someone needs it, I guess |
13 |
- The libcxx{abi,-headers} situation needs love too, since we have |
14 |
different ebuilds for Prefix |
15 |
|
16 |
Fabian |
17 |
|
18 |
On 11-09-2016 02:05:49 +0200, Michael Weiser wrote: |
19 |
> Hi Fabian, |
20 |
> |
21 |
> On Sun, Sep 04, 2016 at 05:06:56PM +0200, Fabian Groffen wrote: |
22 |
> |
23 |
> > I've patched the function to return / in case xcodebuild returns nothing |
24 |
> > useful, such that we fall back to /usr/include, which makes the whole |
25 |
> > thing compile for me. |
26 |
> |
27 |
> Well, I'd like to stress again that having Command Line Tools selected |
28 |
> as the currently selected Xcode does not seem to be supported by the |
29 |
> llvm cmake build system in Apple mode. As far as it's concerned someone |
30 |
> building llvm needs to have a proper Xcode selected using xcode-select |
31 |
> -s so it can find the SDKs within it. |
32 |
> |
33 |
> Is it actually possible to have only Command Line Tools installed |
34 |
> (without Xcode) and get stuff to compile? I'm quite sure it was not in |
35 |
> the olden days (<= 10.6). |
36 |
> |
37 |
> Also, I don't understand why there's a separate instance of the Command |
38 |
> Line Tools somewhere in /Library and how it happens to be the default |
39 |
> "Xcode" for some/most people. And also2, I don't understand why |
40 |
> xcode-select even accepts it as a parameter for -s. This is all very |
41 |
> weird. |
42 |
> |
43 |
> Anyway, I thought of your solution as well while paddling along the |
44 |
> Peene. :) But on second thought I was wondering if we shouldn't patch |
45 |
> any SDK selection and osx-min-version/OSX_DEPLOYMENT_TARGET for the host |
46 |
> system (!) out of llvm/clang for prefix altogether. Prefix is always |
47 |
> meant for the host system, right? Without any of this selection stuff, |
48 |
> the compiler will fall back to the default include paths like |
49 |
> /usr/include anyways and we eliminate a whole lot of possible grief in |
50 |
> the future. IIRC I did this for 3.4.2 for exactly that reason but that |
51 |
> got lost on the move to cmake some time around 3.7, I think. |
52 |
> |
53 |
> The only place I can think of where we'd need SDKs in the context of |
54 |
> prefix is if we wanted to provide a fully cross-capable Apple-like |
55 |
> toolchain for iOS and WatchOS and stuff like that. While binutils-apple |
56 |
> should be able to provide that I guess no-one's ever even tested or |
57 |
> played with it in that direction. And again, we'd need a proper Xcode |
58 |
> installed and selected to find and use the proper SDKs for that. Full |
59 |
> circle... |
60 |
> -- |
61 |
> Thanks, Micha |
62 |
> Ich mag Kaba! |
63 |
> |
64 |
|
65 |
-- |
66 |
Fabian Groffen |
67 |
Gentoo on a different level |