On Sat, Oct 19, 2024 at 4:18 PM Daniel Sahlberg <daniel.l.sahlb...@gmail.com>
wrote:
> Hi,
>
> I'd like to make some changes to tools/dev/unix-build/Makefile.svn but I'd
> like to run then by the community before committing.
>
> Makefile.svn is currently hard coding a dependency version by setting (for
> example)
> RUBY_VER = 2.7.4
> and
> SHA256_${RUBY_DIST} =
> 3043099089608859fc8cce7f9fdccaa1f53a462457e3838ec3b25a7d609fbc5b
> (where RUBY_DIST is derived from RUBY_VER)
>
> I'd like to add several different SHA256_-sums, for "known" dependency
> versions:
> SHA256_ruby-2.7.4.tar.gz =
> 3043099089608859fc8cce7f9fdccaa1f53a462457e3838ec3b25a7d609fbc5b
> SHA256_ruby-2.7.7.tar.gz =
> e10127db691d7ff36402cfe88f418c8d025a3f1eea92044b162dd72f0b8c7b90
> SHA256_ruby-2.7.8.tar.gz =
> c2dab63cbc8f2a05526108ad419efa63a67ed4074dbbcf9fc2b1ca664cb45ba0
> SHA256_ruby-3.0.7.tar.gz =
> 2a3411977f2850431136b0fab8ad53af09fb74df2ee2f4fb7f11b378fe034388
>
> This way it is relatively easy to change a version by changing RUBY_VER to
> one of the supported versions. To support newer versions one would of
> course have to add additional SHA256-sums, but it shouldn't be worse than
> the current situation.
>
I like the idea of saving the SHA256 sums of known versions of
dependencies. It would be a convenience, because currently, if you wish to
use a different version than the default, there is a need to calculate the
sum manually. I've done this a few times. Obviously it will still be
necessary to do this the first time someone uses a version whose sum hasn't
been added yet, but only the first time.
By the way, if the dependency version variables were assigned with '?='
instead of '=', such as:
RUBY_VER ?= 2.7.4
then you could override them at the command line.
*However*, please note that if RUBY_VER happens to be defined in the
environment, the value of the environment variables would take precedence
over the value assigned in the makefile even if not assigned on the command
line. This may or may not be desirable, so using '?=' might be a bad idea.
Personally I'm okay with leaving it as '=' and editing the makefile to
specify the version of a dependency (e.g., RUBY_VER) if a version different
than the default is desired, while saving the SHA256 sums for several known
versions for convenience.
More below:
I'd love to factor out the _VER variables to a separate file that is not
> committed to SVN, best way I could figure out is to let the user create
> Makefile from a new Makefile.tmpl (instead of symlinking to Makefile.svn)
> and including unix-build/Makefile.svn from Makefile(.tmpl). This will be a
> breaking change for thise (few?) using Makefile.svn.
>
I don't understand part of the paragraph above.
I understand the part about factoring out the "known" SHA256 sums to
another file. This could make sense if there are to be myriad SHA256 sums
for all of the different dependencies.
However please note that this makefile is used by creating a symlink to it
(see example usage in adjacent readme) so factoring to a separate file may
require a second symlink or other handling. Unless we're talking about
thousands of lines of SHA256 sums, I'd rather leave it as one file -- but
this is only my opinion given my current understanding (or lack of
understanding) of your idea.
The part I don't understand: Why a file that is *not* committed to SVN?
For Ruby there is an additional complexity that the major/minor version 2.7
> is hardcoded as part of the URL, so I'm planning to split RUBY_VER to two
> separate variables, as is already done for BDB.
>
> Additional Ruby 3 seems to contain the fixes in the separate patch files,
> so I'm planning to make patching optional by:
> [[[
> @@ -1227,8 +1240,10 @@
> tar -C $(SRCDIR) -zxf $(DISTDIR)/$(RUBY_DIST)
> -which ghead && sed -i -e "s/head -c/ghead -c/"
> $(RUBY_SRCDIR)/configure
> $(RUBY_SSL_EX_NEW_DATA_PATCH) $(RUBY_SRCDIR)/ext/openssl/extconf.rb
> +ifeq ($(shell test $(RUBY_MAJOR_VER) \< 3.0; echo 1), 1)
> cd $(RUBY_SRCDIR) && patch -p0 <
> $(RUBY_OBJDIR)/openssl_missing.patch
> cd $(RUBY_SRCDIR) && patch -p0 <
> $(RUBY_OBJDIR)/sparc64_buserror.patch
> +endif
> sed -i 's/X509_num(bs->certs)/X509_num(OCSP_resp_get0_certs(bs))/'
> \
> $(RUBY_SRCDIR)/ext/openssl/ossl_ocsp.c
> touch $@
> ]]]
>
> Serf currently builds without merging the serf/branches/1.3.x-sslbuild
> branch (actually, build fails IF merging). I have a feeling this might be
> needed if using LibreSSL, but I don't have an environment to test it.
>
> Thoughts and comments?
>
I haven't had a chance to grok the rest yet, but wanted to express support
for keeping some known SHA256 sums for convenience (perhaps in a separate
file, perhaps in the same file) and ask why a separate file shouldn't be
committed.
Cheers,
Nathan
