[Codescan] Filter basin sentinel IDs before indexing attractor arrays

[Stardust0831]

Global scan provenance

This issue was found by the Codex global repository scan of non-unit-test files in Stardust0831/Multiwfn, pinned to commit d2b770a531eb7039153198d04520b5992f15c49d.

Problem

Basin generation deliberately stores non-attractor sentinel IDs in gridbas: 0 for unassigned grids and -1 for trajectories reaching an isolated-system box boundary.

Multiwfn/basin.f90

Lines 745 to 753 in d2b770a

	if (ifcorerho==1) then !Restore to valence electron density
	write(*,*) "Restoring valence density..."
	cubmat(:,:,:)=cubmat(:,:,:)-corerhogrid(:,:,:)
	end if
	numunassign=count(gridbas(ixlow:ixup,iylow:iyup,izlow:izup)==0)
	write(*,"(' The number of unassigned grids:',i12)") numunassign
	numgotobound=count(gridbas(ixlow:ixup,iylow:iyup,izlow:izup)==-1)
	write(*,"(' The number of grids travelled to box boundary:',i12)") numgotobound

Multiwfn/basin.f90

Lines 1072 to 1077 in d2b770a

	!Test if encountered box boundary for isolated system
	if (ifPBC==0.and.(inowx==1.or.inowx==nx.or.inowy==1.or.inowy==ny.or.inowz==1.or.inowz==nz)) then
	do itrjgrid=1,ntrjgrid
	gridbas(trjgrid(1,itrjgrid),trjgrid(2,itrjgrid),trjgrid(3,itrjgrid))=-1
	end do
	exit !Terminate trajectory

Low-value-attractor elimination then uses the raw gridbas value to index atttypelist, whose declared lower bound is 1:

Multiwfn/basin.f90

Lines 1200 to 1237 in d2b770a

	subroutine elimlowvalatt(threslowvalatt,imode)
	use defvar
	use basinintmod
	implicit real*8 (a-h,o-z)
	integer imode
	real*8 threslowvalatt
	integer highvalatt(numatt)
	integer att2att(-2:numatt) !Convert indices of old attractors to new ones
	integer atttypelist(numatt)
	if (imode==1) then
	return
	else
	icounthigh=0
	atttypelist=0
	do iatt=1,numatt !Generate list
	if (abs(cubmat(attgrid(1,iatt),attgrid(2,iatt),attgrid(3,iatt)))>=threslowvalatt) then
	icounthigh=icounthigh+1
	highvalatt(icounthigh)=iatt
	atttypelist(iatt)=1 !high
	end if
	end do
	nlowvalatt=numatt-icounthigh
	if (imode==2) then
	do iz=izlow,izup
	do iy=iylow,iyup
	do ix=ixlow,ixup
	if (atttypelist(gridbas(ix,iy,iz))==0) gridbas(ix,iy,iz)=0
	end do
	end do
	end do
	else if (imode==3) then
	do iz=izlow,izup
	do iy=iylow,iyup
	do ix=ixlow,ixup
	call getgridxyz(ix,iy,iz,rnowx,rnowy,rnowz)
	if (atttypelist(gridbas(ix,iy,iz))==0) then !Find out which grid belongs to insignificant attractors
	shortmaxsqr=1D20

The orbital-composition export has the same class of defect: it assigns irealatt=gridbas(...) and uses that raw value to index 1-based composition arrays.

Multiwfn/basin.f90

Lines 2162 to 2191 in d2b770a

	write(*,"(a)") " -4: Export composition of every basin in every orbital to orbcomp.txt"
	!write(*,"(a)") " -5: Print orbital delocalization index (ODI) for a batch of orbitals"
	read(*,"(a)") c80tmp
	if (c80tmp=="0") then
	if (allocated(frag1)) deallocate(frag1)
	return
	else if (c80tmp=="-1") then
	call showorbinfo(1,nmo)
	else if (c80tmp=="-4") then
	if (ifuncbasin/=1) then
	write(*,"(a)") " Error: This option is meaningful only when the function used to partition the basins is electron density!"
	write(*,*) "Press ENTER button to continue"
	read(*,*)
	cycle
	end if
	comparr=0D0
	ifinish=0
	!$OMP PARALLEL private(ix,iy,iz,irealatt,rnowx,rnowy,rnowz,comparrpriv,orbval) shared(ifinish,comparr) NUM_THREADS(nthreads)
	comparrpriv=0D0
	!$OMP do schedule(DYNAMIC)
	do iz=izlow,izup
	do iy=iylow,iyup
	do ix=ixlow,ixup
	call getgridxyz(ix,iy,iz,rnowx,rnowy,rnowz)
	call orbderv(1,1,nmo,rnowx,rnowy,rnowz,orbval(:))
	irealatt=gridbas(ix,iy,iz)
	comparrpriv(:,irealatt)=comparrpriv(:,irealatt)+orbval(:)**2
	end do

Impact

Any grid containing unassigned, boundary, or other non-positive sentinel cells can trigger out-of-bounds reads/writes when these menu paths are used. With bounds checking this can terminate immediately; optimized builds may silently corrupt memory or attribute values to unrelated basins.

Suggested direction

• Treat gridbas <= 0 as an explicit non-attractor case before every attractor-array lookup.
• Preserve or deliberately remap each sentinel instead of passing it through 1-based arrays.
• Centralize basin-ID validation/conversion in a helper so analysis paths cannot repeat raw indexing.
• Add bounds-checked regressions containing -1 and 0 cells for low-value elimination modes 2/3 and orbital-composition export.