Windows build failing in shp2pgsql: XError in psSHP->sHooks.FWrite()

[Algunenano]

From ​https://winnie.postgis.net:444/job/PostGIS_EDB_Regress_winnie/10146/

There are multiple similar issues:

../loader/Point ........ failed (dumper geog test: dumping loaded table: /projects/postgis/tmp/3.1.0dev_pg12_geos3.8.0_gdal2.4.3w64/dumper.err)
-----------------------------------------------------------------------------
Initializing... 
Done (postgis major version: 3).
Output shape: Point
Dumping: XError in psSHP->sHooks.FWrite() while writing object of 28 bytes to .shp file: No error
Error writing shape 0
-----------------------------------------------------------------------------
 ../loader/PointM ... failed ( wkt test: running shp2pgsql output: /projects/postgis/tmp/3.1.0dev_pg12_geos3.8.0_gdal2.4.3w64/loader.err)
-----------------------------------------------------------------------------

By looking at trunk builds, it seems it started failing at:

Changes
Update the shapelib copy to the latest. References #4591. Required two (detail)
	
Generic Cause

	Revision: e7c352608c6449ef59d0adad095289038ba27343
origin/master

Which makes sense since shp2pgsql and shapelib are related.

[robe]

related to #4591

[pramsey]

I'll roll it back out if necessary, I don't have a way to debug this, it's going to have to go into the debugger in windows land.

[robe]

I'll try to debug this in coming weeks, so leave it alone for now. As long as you don't backport to 3.0 we'll figure it out before release time.

[robe]

Okay as the error says "No Error". It is failing

Dumping: XError in psSHP->sHooks.FWrite() while writing object of 28 bytes to .shp file: No error
Error writing shape 0

The issue is in the shpopen.c file. If I swap with the older it works fine.

Also seems to work fine if I change this line

if ( psSHP->sHooks.FWrite( pabyRec, nRecordSize, 1, psSHP->fpSHP ) ) < 1

to

if ( psSHP->sHooks.FWrite( pabyRec, nRecordSize, 1, psSHP->fpSHP ) ) < 0

it's returning 0 so is being sent into that error loop

[robe]

actually I may be mistaken. about it working with that since the regression tests still seem to be failing - though it is outputing to .shp file now. I'm going to compare between old and new outptu to see wha tis failing. I suspect its a maybe the bytes allocated for writing isn't enough.

[Regina Obe <lr@…>]

In 03af80c/git:

Remark out additional guard causing mingw64 to fail. References #4603

[robe]

Remarking out the extra guard condition

    if ( psSHP->sHooks.FTell( psSHP->fpSHP ) != nRecordOffset ) {

seems to have fixed mingw64 issue, but haven't investigated what that extra condition is doing to trigger the issue. — e.g. if it's preventing it from falling into the nested if or is causing something extra to be written to the shp. So I'll leave this open for now until I investigate further.

[robe]

The fix I did I think works. Why it works I have no clue but didn't seem to cause issues for anyone else.