On Jul 12, 2018, at 4:06 PM, Richard Hipp <d...@sqlite.org> wrote:
> On 7/12/18, Jungle Boogie <jungleboog...@gmail.com> wrote:
>> openBSD -current x64
> I don't have access to such a system for debugging purposes.  Can you
> suggest a patch?

I’d suggest revisiting the decision to replace cc-check-function-in-lib with 
the custom check-function-in-lib variant.  The logic behind it seems suspect: 
you *want* it to modify LIBS because you order the calls to the function in 
leaf-to-root order.

That is, the checks should be ordered so:

cc-check-function-in-lib dlopen dl
cc-check-function-in-lib iconv iconv
cc-check-function-in-lib inflateEnd z
cc-check-function-in-lib gethostbyname nsl
cc-check-function-in-lib ns_name_uncompress resolv
cc-check-function-in-lib sqlite3_open sqlite3 $extralibs
cc-check-function-in-lib dlopen dl 

That call list is based on the current contents of auto.def, but consider it 
pseudocode, not a patch, intended just to show the pattern: the most 
depended-upon library is checked for first, because it may be needed to link 
one or more of those that follow, especially on non-Linux systems which usually 
have linkers that won’t chase dependencies for you automatically.

The stock version of cc-check-function-in-lib *prepends* each subsequent 
library for this very reason: so that the first-checked library ends up at the 
*end* of LIBS.
fossil-users mailing list

Reply via email to