Cygwin hooks for stdenvNative and cygwin fixes for various packages #6101
Conversation
|
Could you please rebase this branch relative to the current |
|
@peti i'll do it during this week |
|
So much to review. Maybe this can be splitted in multiple commits? Don't know. |
| @@ -0,0 +1,1710 @@ | |||
| diff -ur w3m-0.5.3/config.guess new/w3m-0.5.3/config.guess | |||
There was a problem hiding this comment.
Was this just generated by autoconf? Might be simpler (and easier to review) to just call autoconf as part of the build, if that doesn't break stuff
|
Some of the .patch files in there are huge. Unless you made all the changes yourself (in which case we should probably look at them very carefully), It seems like it would be cleaner to just refer to wherever the new versions of the source files come from. |
| @@ -255,6 +255,7 @@ stdenv.mkDerivation rec { | |||
| ''; | |||
|
|
|||
| preInstall = "mkdir -p $out/etc/vim"; | |||
| makeFlags = if stdenv.isCygwin then "DESTDIR=/." else null; | |||
There was a problem hiding this comment.
Is there an underlying issue with make on cygwin or is it just asciidoc that has problems?
There was a problem hiding this comment.
It's only asciidoc. Without this "fix" it produces a "//" at the beginning of the path which on cygwin is translated to "" which is the beginning of network SAMBA path.
|
@copumpkin: The patches are mostly taken from cygwin tarballs, i.e. there are no direct urls to them. |
|
@chaoflow @copumpkin what if we create derivation for those patches and extract patches ? it should be possible to refer to patches easier to keep them up to date. |
chaoflow
force-pushed
the
cygwin
branch
3 times, most recently
from
067fb41
to
b12991a
Compare
| @@ -8,7 +8,7 @@ let | |||
| gitBase = lib.makeOverridable (import ./git) { | |||
| inherit fetchurl stdenv curl openssl zlib expat perl python gettext gnugrep | |||
| asciidoc xmlto docbook2x docbook_xsl docbook_xml_dtd_45 libxslt cpio tcl | |||
| tk makeWrapper subversionClient gzip; | |||
| tk makeWrapper subversionClient gzip libiconvOrNull; | |||
There was a problem hiding this comment.
Note that libiconvOrNull will probably go away soon: #6166
|
Apart from the "sed-i" issue, non-cygwin impact seems relatively safe, but it's still a mass rebuild, and it will be better to screen this for larger build regressions before merging to staging. As for the Travis error, it sounds like an escaping issue unrelated to this. We have |
|
Patches: as suggested, on your place I'd consider creating a derivation that extracts all the cygwin patches, so it's easier to update (once upon a time); then they could be used like |
|
Is this still alive? It seems exciting! |
|
@copumpkin still alive. i hope to find time to finish this soon. |
chaoflow
force-pushed
the
cygwin
branch
7 times, most recently
from
8ee83ce
to
ee365e9
Compare
'//' on cygwin is the beginning of a SAMBA path
1. Fix linking against ncurses.dll 2. Fix segfaults with threaded code
|
"if" was used instead of "stdenv.lib.optional/stdenv.lib.optionals/stdenv.lib.optionalString" to not trigger stdenv rebuild. |
|
I see, but for the sake of readability... on staging we do a stdenv rebuild almost every week. |
|
@vcunat this PR is being here for almost half a year. it is really hard to keep it up to date because of patches changing for every version. i would really like to merged it as is, otherwise we would lose what we have currently working. |
|
OK, I thought the PR is waiting for you to finish something
|
|
it is an old comment @chaoflow and I rebased it 2 days ago. |
|
manually merged here |
|
I see; github doesn't notify participants when you add or change PR commits. |
|
Meanwhile, we got around to publish the report: http://ternaris.com/lab/nix-on-windows.html |
Picked from NixOS#6101. (cherry picked from commit e82e14e)
Todo: