summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorEkaitz Zarraga <ekaitz@elenq.tech>2024-04-05 14:35:02 +0200
committerEkaitz Zarraga <ekaitz@elenq.tech>2024-04-05 14:36:31 +0200
commitc83e76374fba54ca2694e5013e0eab4ecf7a38a2 (patch)
tree8037eef9a6b0726c62e3f24477ead80030590ed7
parent3550be337199ca3f872b7074d69c03e59a515c22 (diff)
Update bootstrapGcc12 with a working gcc7.5HEADmaster
-rw-r--r--content/bootstrapGcc/12_gcc_backported_works.md193
1 files changed, 193 insertions, 0 deletions
diff --git a/content/bootstrapGcc/12_gcc_backported_works.md b/content/bootstrapGcc/12_gcc_backported_works.md
new file mode 100644
index 0000000..9747c41
--- /dev/null
+++ b/content/bootstrapGcc/12_gcc_backported_works.md
@@ -0,0 +1,193 @@
+Title: GCC 4.6.4 with RISC-V support in Guix
+Date: 2024-04-02
+Modified: 2024-04-05
+Category:
+Tags: Bootstrapping GCC in RISC-V
+Slug: bootstrapGcc12
+Lang: en
+Summary:
+ We built GCC 4.6.4 with RISC-V support and C++ and all that and we made it
+ run in Guix, finally.
+
+Short post now, following what we did in [previous
+episodes](/bootstrapGcc10.html) where we explained how to build it and
+run it in Debian, now we can run it in Guix thanks to Efraim Flashner's help.
+
+We didn't get rid of the GCC Bootstrapping issue we explained in that post
+(comparing binaries and all that, you remember). But we patched around it
+because the differences were minimal, and this is not going to be a production
+compiler after all. If this causes trouble in the future, we'll fix it, but I
+don't think it will (I've said this before...).
+
+### How to use this thing
+
+We provide a `guix.scm` file in [the repository][gcc] you can use for your own
+packages. If you know Guix, that's probably all you need to know.
+
+[gcc]: https://github.com/ekaitz-zarraga/gcc
+
+In my case, I did some ugly testing on this thing, so I made another folder,
+adjacent to the `gcc` project folder and made this[^ugly]:
+
+[^ugly]: Oh yes, I said ugly. If you know Guix you'll understand what I mean.
+ There are plenty of ways to do this clean, but I don't care. I'm ugly.
+
+I added a `guix.scm` file:
+
+``` scheme
+(use-modules (gnu packages base)
+ (gnu packages)
+ (guix licenses)
+ (guix gexp)
+ (guix utils)
+ (guix build-system gnu)
+ (guix packages))
+
+(load "../gcc/guix.scm")
+
+(define-public my-hello
+ (package
+ (inherit hello)
+ (source (local-file (string-append (dirname (current-filename)) "/hola.cpp")
+ #:recursive? #t))
+ (inputs (list gcc-mine))
+ (arguments
+ `(#:tests? #f
+ #:phases (modify-phases %standard-phases
+ (delete 'configure)
+ (delete 'install)
+ (replace 'build (lambda* (#:key outputs #:allow-other-keys)
+ (let ((out (assoc-ref outputs "out")))
+ (mkdir-p (string-append out "/bin"))
+ (invoke "g++" "--version")
+ (invoke "g++" "hola.cpp" "-o" (string-append out "/bin/hola"))))))))))
+my-hello
+```
+
+And a `hola.cpp` file:
+
+``` clike
+#include<iostream>
+int main (){
+ std::cout << "Hola!\n";
+ return 0;
+}
+```
+
+Jumped into the folder and called this:
+
+``` bash
+guix build -f guix.scm --system=riscv64-linux-gnu -K
+```
+
+Which gave me something like this, but way longer (beware of the `BLABLABLAH`):
+
+``` something
+$ guix build -f guix.scm --system=riscv64-linux -K
+fatal: not a git repository (or any parent up to mount point /)
+Stopping at filesystem boundary (GIT_DISCOVERY_ACROSS_FILESYSTEM not set).
+substitute: updating substitutes from 'https://substitutes.nonguix.org'... 100.0%
+substitute: updating substitutes from 'https://ci.guix.gnu.org'... 100.0%
+substitute: updating substitutes from 'https://bordeaux.guix.gnu.org'... 100.0%
+The following derivation will be built:
+ /gnu/store/niszn6sh1hd6hqwg0xjcw2q7q4lbjvbb-hello-2.12.1.drv
+building /gnu/store/niszn6sh1hd6hqwg0xjcw2q7q4lbjvbb-hello-2.12.1.drv...
+
+... BLABLABLAH ...
+
+starting phase `build'
+g++ (GCC) 4.6.4
+Copyright (C) 2011 Free Software Foundation, Inc.
+This is free software; see the source for copying conditions. There is NO
+warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+
+phase `build' succeeded after 1.4 seconds
+successfully built /gnu/store/niszn6sh1hd6hqwg0xjcw2q7q4lbjvbb-hello-2.12.1.drv
+The following graft will be made:
+ /gnu/store/hnvji18y322q408x3h6i0wafmd6hqmdf-hello-2.12.1.drv
+applying 2 grafts for hello-2.12.1 ...
+
+... BLABLABLAH ...
+
+successfully built /gnu/store/hnvji18y322q408x3h6i0wafmd6hqmdf-hello-2.12.1.drv
+/gnu/store/dbkz567v66ry813svva2j3lnmdchb5r5-hello-2.12.1
+```
+
+And that I could run (using Qemu binfmt magic):
+
+``` bash
+$ /gnu/store/dbkz567v66ry813svva2j3lnmdchb5r5-hello-2.12.1/bin/hola
+Hola!
+```
+
+Let's use `file` command just in case:
+
+``` bash
+$ file /gnu/store/dbkz567v66ry813svva2j3lnmdchb5r5-hello-2.12.1/bin/hola
+/gnu/store/dbkz567v66ry813svva2j3lnmdchb5r5-hello-2.12.1/bin/hola: ELF 64-bit
+LSB executable, UCB RISC-V, RVC, double-float ABI, version 1 (SYSV),
+dynamically linked, interpreter
+/gnu/store/fpqqyym3w1ym24jr2xiwz94hjfgr5hjm-glibc-2.35/lib/ld-linux-riscv64-lp64d.so.1,
+for GNU/Linux 2.6.32, stripped
+```
+
+No tricks here, it's a RISC-V file, and you can see up in the Guix output it
+was built using GCC 4.6.4. That's what we promised, isn't it?
+
+### Next
+
+Next we need to connect this thing to the bootstrapping chain. It needs to be
+built using TinyCC and it needs to be able to build a modern GCC that is more
+stable.
+
+With that, we might be able to build the whole RISC-V world.
+
+Wish us luck.
+
+
+#### Extra (added 2024-04-05)
+
+In a similar fashion to the `my-hello` I shared as an example, I just tried
+this:
+
+``` scheme
+(use-modules (gnu packages gcc)
+ (ice-9 rdelim)
+ (gnu packages)
+ (guix licenses)
+ (guix gexp)
+ (guix utils)
+ (guix build-system gnu)
+ (guix packages))
+
+(load "../gcc/guix.scm")
+
+(define-public gcc-with-mine
+ (package
+ (inherit gcc-7)
+ (inputs `(("gcc" ,gcc-mine) ,@(alist-delete "gcc" (package-inputs gcc-7))))
+ (arguments (substitute-keyword-arguments (package-arguments gcc-7)
+ ((#:phases phases)
+ `(modify-phases ,phases
+ (add-after 'unpack 'setenv
+ ;; We don't want gcc-11:lib in CPLUS_INCLUDE_PATH, it messes with
+ ;; libstdc++ from gcc-4.6.
+ (lambda _
+ (setenv "CPLUS_INCLUDE_PATH" (getenv "C_INCLUDE_PATH")))) ))))))
+
+gcc-with-mine
+```
+
+And guess what? It worked.
+
+This is really interesting because GCC 7.5 is the first one with official
+RISC-V support. Being able to build it means from here we can follow the
+existing bootstrapping chain and build the world. Also, GCC is a huge program
+and, as I told you before, it does some checks during the build, building
+itself twice and then comparing the resulting binaries to make sure it's
+stable. Building an stable GCC that is also able to build itself means a lot.
+
+At this point we can consider my GCC-4.6.4 backport is correct, and not only
+that, our package definition is, too. In projects like this bootstrapping thing
+it still feels weird when things just work, it makes you be suspicious... I'll
+try to ignore that feeling and be happy, just for once.