Thursday, 28 March 2019

Announcing: MapGuide Open Source 3.1.2

I am pleased to announce the availability of MapGuide Open Source 3.1.2

The notable changes in this release are detailed below:

Updated Web Tier Components

MapGuide Open Source bundles updated web tier components:

  • Apache HTTPD 2.4.37
  • Tomcat 7.0.92
  • PHP 5.6.40
Since PHP 5.6 is now end-of-life, this is most likely the last release in the 3.1 branch as we work to bring PHP 7 support for the MapGuide API in the next major release.

Functional SHP FDO provider for 64-bit Linux

Over the new year, I had set out to solve one of the last remaining blockers that prevented us from having a functional 64-bit SHP FDO provider for 64-bit Linux.

I had speculated that the blocker is due to incorrect data size assumptions in the SHP spatial indexing code on 64-bit Linux which would throw off spatial index queries. Intensive debugging sessions proved this to be the case and I was finally able to land the long-awaited fixes which is now available with the SHP provider bundled with this release.

The proof that the provider was truly functional was being able to load a SHP-heavy dataset into a 64-bit MapGuide Server and being able to pan, zoom, select and buffer SHP features without any errors or features disappearing in and out at various zoom levels (due to previously out-of-whack SHP spatial index queries)

Experimental King.Oracle FDO provider for Linux (32 and 64-bit)

Since I got a major win with one Linux FDO blocker and was on a roll, I figured we can do a double-whammy with the other major Linux FDO blocker: The lack of a functional King.Oracle FDO provider on Linux.

For this one I had speculated that the problem was due to heavy usage of wchar_t for strings which (due to different size assumptions on Linux) meant that most strings going in and out of the provider on Linux were garbage, meaning the SQL queries we think we were sending to OCI for execution were completely scrambled.

Due to the lack of solid examples of using OCI with wchar_t strings on Linux, I figured the code as it currently stands will never work on Linux. So I took the page out of a reference implementation I knew to definitely work on Linux (the OCI driver for GDAL/OGR) and wholesale #ifdef'd the entire provider codebase to use narrow (char) strings for Linux and their expected OCI functions, while retaining wide (wchar_t) strings and current OCI function usage for Windows.

The ability to easily spin up an Oracle XE environment (well until recently at least) meant that I was able to quickly verify that switching to narrow strings on Linux was indeed the way to go. This also provided motivation for me to implement a basic (and repeatable) unit test suite for this provider, finally giving us an objective baseline level of validation and verification for this provider.

The end result of all this is that the King.Oracle provider now not only builds on Linux, but it now finally works!

However, I'm certain that there will be bugs in this provider for Linux waiting to be revealed, so the provider gets the "experimental" label. Use at your own risk/discretion. Please do report any issues you find.

OGR FDO Provider Improvements

The OGR provider has undergone many "quality of life" improvements:
  • We've cleaned up various correctness issues around provider behavior. For example, attempting to access invalid property names in an OGR feature reader will now throw instead of silently failing before.
  • The provider supports a new DefaultSchemaName property that lets you specify a default schema name other than OGRSchema
  • The provider also supports a new DataSourceEncoding property to allow you declare the encoding for an OGR data source where it cannot be inferred by default by the OGR library so that properties/names/etc with non-english characters are rendered properly.
Fusion Changes
Other Changes

No comments: