Re: [HACKERS] ICU support on Windows

On Tue, Jun 13, 2017 at 6:47 AM, Alvaro Herrera
<alvherre@2ndquadrant.com> wrote:
> Michael Paquier wrote:
>> I do some low level packaging on Windows (libxml2, libxslt, etc.), and
>> the compilation code usually allows you to usually use the
>> installation paths you want. At the end using only lib/ looks more
>> generic to me, and I did the same renaming as Ashutosh after unzipping
>> their files. There is already "Program Files" and "Program Files
>> (x86)" to make such distinctions.
>
> Oh my.  And then they say Microsoft has the brightest minds in the
> planet ...  (apparently they're all at Facebook nowadays actually. Go
> figure.)

:)

> Is this a problem on the packaging machine only, or does it cause
> fall-out on the machine in which the program runs?

From Windows point of view, it depends on if you use /MD or /MT with cl.exe:
https://msdn.microsoft.com/en-us/library/2kzt1wy3(v=vs.110).aspx
/MD makes binaries dynamically link to DLLs, making binaries sensitive
to system updates. While /MT makes a binary bigger as the library
becomes embedded into it.

It is possible to see what an exe needs at runtime by using dumpbin
/imports which lists all the dependencies needed by the binaries. And
one can notice which option is used with cl.exe:
- If msvcrXXX.dll is listed (XXX is the MSVC version), then /MD was used.
- If msvcrXXXd.dll is listed, then /MDd was used.
- If nothing is listed, then /MT was used.
IMO, the places where the dll are placed is the problem of the
packager, and Postgres uses /MD so if dll are misplaced there will be
a run-time error because dependencies cannot be loaded.
-- 
Michael