[Debian-med-packaging] Bug#1153516: arb: FTBFS on 32-bit architectures: arb_progress(const char*, unsigned int) is private
Cheng Li
shiptux at gmail.com
Mon Nov 9 12:32:00 GMT 2026
Source: arb
Version: 7.0+ds-1
Severity: serious
Tags: ftbfs patch
Dear Maintainer,
arb 7.0+ds-1 fails to build from source on i386:
https://buildd.debian.org/status/fetch.php?pkg=arb&arch=i386&ver=7.0%2Bds-1&stamp=1792765920&raw=0
The visible failure is a link error:
/usr/bin/i686-linux-gnu-ld.bfd: cannot find adname.o: No such file or directory
/usr/bin/i686-linux-gnu-ld.bfd: cannot find TreeNode.o: No such file or directory
make[5]: *** [Makefile:89: ../lib/libARBDB.so] Error 1
but those objects are missing because their compilation failed earlier in the
same parallel build:
adname.cxx:181:131: error: 'arb_progress::arb_progress(const char*, NUM)
[with NUM = unsigned int]' is private within this context
CORE/arb_progress.h:253:38: note: declared private here
There are six such call sites in three files:
ARB_GDE/GDE_event.cxx:363 -> breaks ARB_GDE.a
ARBDB/adname.cxx:181, 219, 272 -> breaks libARBDB.so
ARBDB/TreeNode.cxx:425, 458 -> breaks libARBDB.so
This is not specific to i386, it is specific to 32-bit. CORE/arb_progress.h
makes a template constructor private on purpose, to reject integer types that
could overflow, and then explicitly overloads the wanted ones:
private:
// forbid constructing arb_progress with integer types that could overflow in caller code:
template <typename NUM> arb_progress(const char *t, NUM n);
template <typename NUM> explicit arb_progress(NUM n);
public:
// explicitely overload wanted ctors:
explicit arb_progress(char *title) { setup(title, 0); }
arb_progress(const char *title, unsigned long overall_count) { ... }
explicit arb_progress(unsigned long overall_count) { ... }
The only public constructor taking a count accepts unsigned long, and all six
call sites pass a size_t:
adname.cxx:181 tree_count*3 (printed with %zu)
adname.cxx:219 config_count (printed with %zu)
adname.cxx:272 colorset_count (printed with %zu)
TreeNode.cxx:425 eliminatedParentLength.size()
TreeNode.cxx:458 eliminatedParentLength.size()
GDE_event.cxx:363 dataset.numelements-oldnumelements+1
(NA_Alignment::numelements is size_t, GDE_def.h:120)
On LP64 architectures size_t is unsigned long, so the public overload is an
exact match and the code compiles. On ILP32 architectures size_t is unsigned
int, the private template instantiated with NUM = unsigned int is then a
better match than the unsigned long overload, and overload resolution selects
the private one.
arb 6.0.6-8 still built successfully on i386, so this is a regression in
7.0+ds-1.
The patch below adds the matching unsigned int overload, mirroring the
existing unsigned long one:
--- a/CORE/arb_progress.h
+++ b/CORE/arb_progress.h
@@ -256,6 +256,7 @@
// explicitely overload wanted ctors:
explicit arb_progress(char *title) { setup(title, 0); }
arb_progress(const char *title, unsigned long overall_count) { arb_assert(overall_count<=LONG_MAX); setup(title, overall_count); }
+ arb_progress(const char *title, unsigned int overall_count) { arb_assert(overall_count<=LONG_MAX); setup(title, overall_count); }
explicit arb_progress(unsigned long overall_count) { arb_assert(overall_count<=LONG_MAX); setup(NULp, overall_count); }
void allow_title_reuse() { used->allow_title_reuse(); }
I verified the mechanism and the fix with a minimal reproducer of just that
constructor set (g++ 14.2.0-19, no arb headers involved):
g++ -m64 -> OK
g++ -m32 -> is private within this context
g++ -m32 -DWITH_FIX -> OK
g++ -m64 -DWITH_FIX -> OK
so the added overload fixes 32-bit without changing 64-bit behaviour. I have
not run a full arb build on i386, so the patch is verified at the overload
resolution level only.
Two further notes:
- armhf is affected by the same bug, but it currently fails earlier on the
bundled RAxML 8.2.3 and -msse3 (#1153055), so this one is masked there.
Fixing #1153055 will expose it.
- For symmetry the single-argument "explicit arb_progress(unsigned int)" could
be added as well, but no current call site needs it, so the patch leaves it
out.
Kind regards,
Cheng Li
More information about the Debian-med-packaging
mailing list