__time64_t and 32bit platform

[yopito]

while packaging passwordsafe 1.08.1BETA, build fails for 32bit platform.

Context: VoidlLinux distribution with gcc 9.1 and glibc 2.29 or musl libc 1.1.22

Example of fail build: https://travis-ci.org/void-linux/void-packages/jobs/535496599e

Offending starts at is https://github.com/pwsafe/pwsafe/blob/master/src/os/unix/pws_time.h#L18

my points:

• "__time64_t" typedef is defined on 32bit platform, even if "__time64_t" define statement is not
• fill confused that "__time64_t" typedef is defined in pws_time.h, that is a structure defined in libc headers (see [1])
• libc header (see [1]) also states that "__time64_t" is always a 64bit type (via __TIME64_T_TYPE)

Fixed myself build via replacing L18 with
#if !defined(__time64_t) && !defined(__TIME64_T_TYPE)
It's working, but I'm not sure this is the right way to fix it

So pws_time.h should not redefine __time64_t or use another name for that ?
I'm not enough fluent in C/C++ to propose a good patch, sorry.

[1] Gnu libc include/bits/time64.h states:

26  /* Define __TIME64_T_TYPE so that it is always a 64-bit type.  */
27
28  #if __TIMESIZE == 64
29  /* If we already have 64-bit time type then use it.  */
30  # define __TIME64_T_TYPE                __TIME_T_TYPE
31  #else
32  /* Define a 64-bit time type alongsize the 32-bit one.  */
33  # define __TIME64_T_TYPE                __SQUAD_TYPE
34  #endif
35
36  #endif /* bits/time64.h */

[ronys]

Fixed in commit 497246d. I took your fix, and removed the unused __time32_t typedef.

It's not that pwsafe is redefining __time64_t, but rather defining it for environments where it's missing, making the rest of the code more consistent.

[yopito]

cool, quick response.
FYI, with your fix, build with various arch and libc combinations is fine (voidlinux).
See https://travis-ci.org/void-linux/void-packages/builds/554874381 for instance

[ronys]

Nice. Interesting to see builds for ARM as well.

Care to submit a pull request to integrate voidlinux support into the main repo?

[yopito]

Thanks for your interrest to VoidLinux, but I don't well understand how it could be achieved in pwsafe context: specific entry in CmakeList.txt ? Voidlinux as travis target ?

[ronys]

• Well, if you made any changes to the source code besides the fix we've discussed, that would be of interest.
• If cmake supports voidlinux packages, then changes to CmakeList.txt would be relevant
• If the voidlinux package can be built by a script on a voidlinux system, then that script can be added under the install directory in the project repo.
• Adding voidlinux to the existing travis.yml file would also be useful, I think.

[yopito]

sorry for late reply. It feels difficult to answer you clearly, so here my points.

VoidLinux distributes passwordsafe as an official package:

• source packaging is here: https://github.com/void-linux/void-packages/tree/master/srcpkgs/passwordsafe/
• Currently, there are minor changes against passwordsafe source code to achieve packaging (see "template" file). There will be even less changes in next version.
• Currently in version 1.07BETA, but a PR is opened to update in 1.08.1BETA:
  passwordsafe: update to 1.08.1BETA void-linux/void-packages#11854
  This one contains the patch you wrote about the current PR

To achieve packaging, VoidLinux uses a integrated tree that contains both the source packaging definition of all packages and the shell build utility and its helpers (xbps-src).
So:

• The packaging of a given application has to be maintained in that tree
• I could provide you a copy of the templating file for passwordsafe, but it sounds a bit useless since template directives heavily rely on the build ecosystem that is constantly evolving (this is a rolling distribution). Difficult to maintain ?
• Travis does not offer a VoidLinux image. VoidLinux uses travis on top of Ubuntu image with a static package manager and retrieve the combined tree to launch test packaging. Official build is achieved on dedicated systems of the project. So using VoidLinux as a QA target for passwordsafe can be tricky ?

Anyway, I will be glad to help you on any of these points if you want to move on.

[ronys]

Thanks for clarifying. Based on your explanation, I don't see any real value in adding voidlinux specifics to the main passwordsafe repo.