aboutsummaryrefslogtreecommitdiff
path: root/build2/cxx/link.cxx
diff options
context:
space:
mode:
authorBoris Kolpackov <boris@codesynthesis.com>2016-06-20 18:48:40 +0200
committerBoris Kolpackov <boris@codesynthesis.com>2016-06-20 18:49:17 +0200
commit70d00b9f7f3266c1962f6d5a6fc8de1866c67949 (patch)
tree21f856fff252eb147b7087274f3cb5bc87425d6f /build2/cxx/link.cxx
parentb0e481a653b01e4329bccb1d101d56e3e878e960 (diff)
Fix some more MinGW .dll.a search
Diffstat (limited to 'build2/cxx/link.cxx')
-rw-r--r--build2/cxx/link.cxx58
1 files changed, 43 insertions, 15 deletions
diff --git a/build2/cxx/link.cxx b/build2/cxx/link.cxx
index d085a1d..bf2b992 100644
--- a/build2/cxx/link.cxx
+++ b/build2/cxx/link.cxx
@@ -318,7 +318,7 @@ namespace build2
sn = path ("lib" + p.name);
if (tsys == "darwin") e = "dylib";
- else if (tsys == "mingw32") e = "dll";
+ else if (tsys == "mingw32") e = "dll.a"; // See search code below.
else e = "so";
}
@@ -377,19 +377,32 @@ namespace build2
{
f = d;
f /= sn;
+ mt = file_mtime (f);
- // @@ Here we are searching for the import library but if it's not
- // found, then we could also search for the DLL since we can link
- // directly to it. Of course, we would also need some way to mark
- // this libso{} as "import-less".
- //
- // @@ This is a bit of hack until we get support for the explicit
- // import lib specification.
- //
if (tsys == "mingw32")
- mt = file_mtime (f + ".a");
- else
- mt = file_mtime (f);
+ {
+ // Above we searched for the import library (.dll.a) but if it's
+ // not found, then we also search for the .dll (unless the
+ // extension was specified explicitly) since we can link to it
+ // directly.
+ //
+ // Note also that the resulting libso{} would end up being the .a
+ // import library. We could have tried to always find the
+ // corresponding .dll, but when installed they normally end up in
+ // different directories (lib/ and bin/). In fact, there is no
+ // reason to require the presence of the .dll at all (think cross-
+ // compilation). So while having libso{} being .a is a bit of
+ // hack, it is simple and appears harmless (think of it as the
+ // import library being a proxy for the real thing). But let me
+ // know if you have a better idea.
+ //
+ if (mt == timestamp_nonexistent && ext == nullptr)
+ {
+ se = &extension_pool.find ("dll");
+ f = f.base (); // Remove .a from .dll.a.
+ mt = file_mtime (f);
+ }
+ }
if (mt != timestamp_nonexistent)
{
@@ -1232,15 +1245,30 @@ namespace build2
((ppt = a = pt->is_a<liba> ()) ||
(ppt = so = pt->is_a<libso> ()))))
{
- string p (relative (ppt->path ()).string ()); // string()&&
+ path p (relative (ppt->path ()));
if (so != nullptr)
{
if (tsys == "mingw32")
- p += ".a"; // Link to the import library (*.dll -> *.dll.a).
+ {
+ // Normally we want to link to the import library (*.dll ->
+ // *.dll.a). However, this could already be an import library
+ // (see search_library()).
+ //
+ // @@ It could also be a DLL which we should try to link
+ // directly. We cannot handle it until we have the group
+ // support: we will check if there is an implib in
+ // the group, if there is, then we link to it. Otherwise,
+ // we assumy this is a DLL without the import library and
+ // try to link to it directly.
+ //
+ const char* e (p.extension ());
+ if (e == nullptr || e[0] != 'a' || e[1] != '\0')
+ p += ".a";
+ }
}
- sargs.push_back (move (p));
+ sargs.push_back (move (p).string ()); // string()&&
// If this is a static library, link all the libraries it depends
// on, recursively.