cplusplus: apply a few clang-tidy suggested fixes

[kleisauke]

See the individual commits for more details.

The abi-compliance-checker results are available at:
https://kleisauke.nl/compat_reports/vips-cpp/master_to_clang-tidy/compat_report.html

[kleisauke]

I considered removing what(), since it adds no functional value beyond what the base class (std::runtime_error) already provides. However, doing so appears to break ABI compat, see:
https://kleisauke.nl/compat_reports/vips-cpp/master_to_clang-tidy-rev2/compat_report.html
https://github.com/kleisauke/libvips/compare/cpp-clang-tidy..cpp-clang-tidy-rev2

[jcupitt]

Wow I'm amazed you can change the exception base class without breaking ABI compatibility. The C++ ABI rules are very mysterious.

Is it possible it could break with some C++ compilers and not others?

[kleisauke]

My understanding is that this is safe because std::runtime_error also stores a std::string internally. However, I need to verify this, since the change in class size could(?) break code that catches VError by value (catch (VError err)) rather than by reference (catch (const VError &err)).

[kleisauke]

... let's be on the safe side, commit 943d1fc ensures ABI compatibility, I now see:
https://kleisauke.nl/compat_reports/vips-cpp/master_to_clang-tidy-abi/compat_report.html

[jcupitt]

Great!

[lovell]

I've been investigating error message corruption in v8.18.0 on Windows and I suspect this change might have exposed us to microsoft/STL#2114. The call to vips_error_buffer now takes place sooner at construction time rather than deferring to a later what() call. Perhaps we should use vips_error_buffer_copy and manage the lifetime within the exception?

[kleisauke]

@lovell According to PR nodejs/node-addon-api#1059, this define appears to be the root cause:
https://github.com/nodejs/node/blob/v25.4.0/common.gypi#L485-L486

I'm not sure if we should support the _HAS_EXCEPTIONS=0 case. This changeset appears to resolve the issue in sharp:
lovell/sharp@0.35...kleisauke:sharp:temp-err-2
(i.e. the define also needs to be propagated to consumers of VError8.h)

[lovell]

@kleisauke Thanks, that explains it, I was so close! It looks like STL will be altering its behaviour to copy the message so we can leave this alone.