| Version annotated: | 3.3.2~rc2-2 |
| Comments: |
tl;dr: A floating-point constant has a different value. . Occurs on amd64 and i386 but not on armhf. This accords with architecture-dependent dependencies (no pun intended) in d/control and d/rules, specifically, libfftw3-long3 is not available on armhf. . The library is built three times, with (float | double | long double) as the type 'R'. (Reminder: on amd64, the sizeof()s of these types are 4/8/16 respectively and they must be aligned to their size.) . The difference is always in the .rodata section, always right after the end of a string literal and before what appears to be an array of small integers in 16-bit words; for example, in the 'float' case: . │ │ │ │ │ 0x0004abe0 45252b2e 3145692c 00252b45 252b4569 E%+.1Ei,.%+E%+Ei │ │ │ │ │ 0x0004abf0 2c0a0000 f037983c f304b539 0000f042 ,....7.<...9...B │ │ │ │ │ - 0x0004ac00 e30c1b4f 286b6e4e 00000000 00000000 ...O(knN........ │ │ │ │ │ + 0x0004ac00 53eb264f 286b6e4e 00000000 00000000 S.&O(knN........ │ │ │ │ │ 0x0004ac10 00000000 00000000 00000000 00000000 ................ │ │ │ │ │ 0x0004ac20 01000000 00000000 03000000 00000000 ................ │ │ │ │ │ 0x0004ac30 07000000 00000000 09000000 00000000 ................ │ │ │ │ │ 0x0004ac40 0e000000 00000000 11000000 00000000 ................ . The difference always occurs in the 4th sizeof(R)-block after the end of the string literal "%+E%+Ei,\n". For example, in the 'float' case the literal ends at 0x4abf2, the next point a float can be aligned at is 0x4abf4, and «(float*)(3 + (float*)0x4abf4)» is the word having a difference. . The same is true for the 'double' and 'long double' builds, mutatis mutandis (substitute for 'float' the value of the type 'R' in that build). . Tracing the architecture-dependent part (d/rules → --enable-long-double → AC_DEFINE(NFFT_LDOUBLE, 1)) finds two global arrays of type 'R': . kernel/util/lambda.c|24 col 13| #if defined(NFFT_LDOUBLE) kernel/util/bessel_i0.c|21 col 13| #if defined(NFFT_LDOUBLE) . These arrays would be stored in .rodata, although I couldn't find them in the symbol table. . So, the hypothesis is that a const variable of type 'R' of static storage duration has a different value between the two builds. |
|
Our notes about issues affecting packages are stored in notes.git and are targeted at packages in Debian in 'unstable/amd64' (unless they say otherwise). |
|