prevent invalid conversion from const compile error

[jgrocha]

Compiling PDAL was breaking with:

/home/jgr/dev/cpp/PDAL/pdal/Geometry.cpp: In member function ‘pdal::Utils::StatusWithReason pdal::Geometry::transform(pdal::SpatialReference)’:
/home/jgr/dev/cpp/PDAL/pdal/Geometry.cpp:175:61: error: invalid conversion from ‘const OGRSpatialReference*’ to ‘OGRSpatialReference*’ [-fpermissive]
  175 |     OGRSpatialReference *inSrs = m_geom->getSpatialReference();
      |                                  ~~~~~~~~~~~~~~~~~~~~~~~~~~~^~
      |                                                             |
      |                                                             const OGRSpatialReference*

I'm running on Ubuntu, with:

gcc --version
gcc (Ubuntu 11.3.0-1ubuntu1~22.04) 11.3.0

gdalinfo --version
GDAL 3.7.0dev-7cd1a9bb31, released 2023/04/17 (debug build)

Tests were passed after this change.

[hobu]

picked back to 2.5-maintenance at 516e33a

[dg0yt]

Hm, this change does indeed "prevent invalid conversion from const compile error" messages.
But why choose const_cast which only legalizes the evil conversion, instead of actually accepting the constness returned by getSpatialReference()?
Was this meant to retain compatibility with older versions of gdal where the following methods are not able to use a pointer to const?
Even in that case, isn't PDAL compiled with C++17 so that auto could help accomodate the actual type?

[abellgithub]

I think the type can just be made const and the cast can be dropped. WFM, anyway.

[dg0yt]

I think the type can just be made const and the cast can be dropped.

I already know that this works with gdal 3.7.0. (I'm trying to patch this for the gdal port in vcpkg.)
The question is whether older versions of gdals provide GetRoot etc. for pointers to const.

WFM, anyway.

IMO this is a bad guideline when it comes to const_cast. It is hiding a violating of API contracts. Even if it works today with one compiler, results may be different with another update of gdal, another compiler, or just more aggressive optimization settings. And it will break at runtime, not buildtime.

[abellgithub]

IMO this is a bad guideline when it comes to const_cast. It is hiding a violating of API contracts. Even if it works today with one compiler, results may be different with another update of gdal, another compiler, or just more aggressive optimization settings. And it will break at runtime, not buildtime.

What is a bad guideline? What would break at runtime?

[dg0yt]

What is a bad guideline?

"WFM, anyway" is a bad guideline IMO when it comes to const_cast. Recommended guideline: "Don't cast away const."

What would break at runtime?

Nobody knows. "Modifying a const object through a non-const access path ... results in undefined behaviour.".

---

Repeating my questions in other words:
Was there a concern with regard to old gdal version for preferring const_cast?
Would pdal accept a PR which uses auto?

[abellgithub]

You misundersood. Making the type const, rather than applying the cast should be fine. WFM meant that such a change compiles for me (the pointer to const is allowed in the contexts in which it is used).