Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

grepdiff: --exended-regexp does not work when compiled --with-pcre2 #60

Open
phil-blain opened this issue Mar 5, 2024 · 6 comments
Open

Comments

@phil-blain
Copy link

phil-blain commented Mar 5, 2024

I discovered that if the package is compiled with --with-pcre2 (which is the default behaviour if libpcre2 is found on the system), then the following invocation does not return anything for the diff copied at the end of this message.

grepdiff  --output-matching=hunk -E 'landblockelim|blockall' out.txt

Since either landblockelim or blockall are found in several of the hunks, I should get an output. However, if I use the BRE syntax instead:

grepdiff  --output-matching=hunk -E 'landblockelim\|blockall' out.txt

then I get the expected output.

I ran the code through GDB and I see the regexp is correctly passed to regcomp with REG_EXTENDED set, but in <pcre2posix.h>, which is included instead of <regex.h> under --with-pcre2, we have the following:

/* This is not used by PCRE2, but by defining it we make it easier
to slot PCRE2 into existing programs that make POSIX calls. */

#define REG_EXTENDED  0

so it seems something is missing for a working "out-of-the box" experience when compiling with libpcre2...


The diff in question:

commit 078aab4844983ccbb9c9fddc1a70e060e1d99a34
Author: Tony Craig <[email protected]>
Date:   May 10 2022

    Merge cgridDEV branch including C grid implementation and other fixes (#715)
    
    Merge cgridDEV branch to main
    
diff --git a/cicecore/cicedynB/infrastructure/ice_domain.F90 b/cicecore/cicedynB/infrastructure/ice_domain.F90
index c44e896a..79f5bcb9 100644
--- a/cicecore/cicedynB/infrastructure/ice_domain.F90
+++ b/cicecore/cicedynB/infrastructure/ice_domain.F90
@@ -63,6 +63,8 @@ module ice_domain
       maskhalo_dyn   , & ! if true, use masked halo updates for dynamics
       maskhalo_remap , & ! if true, use masked halo updates for transport
       maskhalo_bound , & ! if true, use masked halo updates for bound_state
+      halo_dynbundle , & ! if true, bundle halo update in dynamics
+      landblockelim  , & ! if true, land block elimination is on
       orca_halogrid      ! if true, input fields are haloed as defined by orca grid
 
 !-----------------------------------------------------------------------
@@ -79,6 +81,7 @@ module ice_domain
                              ! 'rake', 'spacecurve', etc
        distribution_wght     ! method for weighting work per block 
                              ! 'block' = POP default configuration
+                             ! 'blockall' = no land block elimination
                              ! 'latitude' = no. ocean points * |lat|
                              ! 'file' = read distribution_wgth_file
     character (char_len_long) :: &
@@ -153,13 +156,15 @@ subroutine init_domain_blocks
    maskhalo_dyn      = .false.     ! if true, use masked halos for dynamics
    maskhalo_remap    = .false.     ! if true, use masked halos for transport
    maskhalo_bound    = .false.     ! if true, use masked halos for bound_state
+   halo_dynbundle    = .true.      ! if true, bundle halo updates in dynamics
    add_mpi_barriers  = .false.     ! if true, throttle communication
    debug_blocks      = .false.     ! if true, print verbose block information
-   max_blocks        = -1           ! max number of blocks per processor
+   max_blocks        = -1          ! max number of blocks per processor
    block_size_x      = -1          ! size of block in first horiz dimension
    block_size_y      = -1          ! size of block in second horiz dimension
    nx_global         = -1          ! NXGLOB,  i-axis size
    ny_global         = -1          ! NYGLOB,  j-axis size
+   landblockelim     = .true.      ! on by default
 
    if (my_task == master_task) then
       write(nu_diag,*) subname,' Reading domain_nml'
@@ -284,7 +289,7 @@ end subroutine init_domain_blocks
 
 !***********************************************************************
 
- subroutine init_domain_distribution(KMTG,ULATG)
+ subroutine init_domain_distribution(KMTG,ULATG,grid_ice)
 
 !  This routine calls appropriate setup routines to distribute blocks
 !  across processors and defines arrays with block ids for any local
@@ -299,6 +304,9 @@ subroutine init_domain_distribution(KMTG,ULATG)
       KMTG           ,&! global topography
       ULATG            ! global latitude field (radians)
 
+   character(len=*), intent(in) :: &
+      grid_ice         ! grid_ice, B, C, CD, etc
+
 !----------------------------------------------------------------------
 !
 !  local variables
@@ -316,6 +324,7 @@ subroutine init_domain_distribution(KMTG,ULATG)
    integer (int_kind) :: &
       i,j,n              ,&! dummy loop indices
       ig,jg              ,&! global indices
+      igm1,igp1,jgm1,jgp1,&! global indices
       ninfo              ,&! ice_distributionGet check
       work_unit          ,&! size of quantized work unit
 #ifdef USE_NETCDF
@@ -449,6 +458,8 @@ subroutine init_domain_distribution(KMTG,ULATG)
        flat = 1
    endif
 
+   if (distribution_wght == 'blockall') landblockelim = .false.
+
    allocate(nocn(nblocks_tot))
 
    if (distribution_wght == 'file') then
@@ -504,10 +515,25 @@ subroutine init_domain_distribution(KMTG,ULATG)
                   if (this_block%i_glob(i) > 0) then
                      ig = this_block%i_glob(i)
                      jg = this_block%j_glob(j)
-                     if (KMTG(ig,jg) > puny .and.                      &
-                        (ULATG(ig,jg) < shlat/rad_to_deg .or.          &
-                         ULATG(ig,jg) > nhlat/rad_to_deg) )            & 
-                          nocn(n) = nocn(n) + flat(ig,jg)
+                     if (grid_ice == 'C' .or. grid_ice == 'CD') then
+                        ! Have to be careful about block elimination with C/CD
+                        ! Use a bigger stencil
+                        igm1 = mod(ig-2+nx_global,nx_global)+1
+                        igp1 = mod(ig,nx_global)+1
+                        jgm1 = max(jg-1,1)
+                        jgp1 = min(jg+1,ny_global)
+                        if ((KMTG(ig  ,jg  ) > puny .or.                             &
+                             KMTG(igm1,jg  ) > puny .or. KMTG(igp1,jg  ) > puny .or. &
+                             KMTG(ig  ,jgp1) > puny .or. KMTG(ig  ,jgm1) > puny) .and. &
+                            (ULATG(ig,jg) < shlat/rad_to_deg .or.          &
+                             ULATG(ig,jg) > nhlat/rad_to_deg) )            & 
+                             nocn(n) = nocn(n) + flat(ig,jg)
+                     else
+                        if (KMTG(ig,jg) > puny .and.                      &
+                           (ULATG(ig,jg) < shlat/rad_to_deg .or.          &
+                            ULATG(ig,jg) > nhlat/rad_to_deg) )            & 
+                             nocn(n) = nocn(n) + flat(ig,jg)
+                     endif
                   endif
                end do
             endif
@@ -524,8 +550,8 @@ subroutine init_domain_distribution(KMTG,ULATG)
          ! Keep all blocks even the ones only containing land points
          if (distribution_wght == 'block') nocn(n) = nx_block*ny_block
 #else
-         if (distribution_wght == 'block' .and. &   ! POP style
-             nocn(n) > 0) nocn(n) = nx_block*ny_block
+         if (distribution_wght == 'block' .and. nocn(n) > 0) nocn(n) = nx_block*ny_block
+         if (.not. landblockelim) nocn(n) = max(nocn(n),1)
 #endif
       end do
    endif  ! distribution_wght = file
@phil-blain
Copy link
Author

phil-blain commented Mar 6, 2024

So, after more investigation, this behaviour only appears on RHEL8, with a self-compiled patchutils 0.4.2 using the system-installed libpcre2 (pcre2-10.32-2.el8.src.rpm).

It does not appear:

  • On Ubuntu 22 with a self-compiled patchutils 0.4.2 using the system-installed libpcre2 (10.39-3ubuntu0.1)
  • On Ubuntu 22 with a self-compiled patchutils 0.4.2 using a self-compiled pcre2 at the same version as on RHEL 8 (10.32)

@phil-blain
Copy link
Author

Turns out, the system where I was seeing this behaviour had LD_PRELOAD=/path/to/libc.so set in the environment, which caused the libc regcomp and regexec functions to be linked at runtime, which explains why I had to use the BRE alternation syntax \|. And since prec2posix.h defines REG_EXTENDED=0, using -E had no effect.

@phil-blain
Copy link
Author

Note that this would not have happened with pcre2 10.33 or later since the POSIX function names are only defined as preprocessor macros starting in that release: https://github.com/PCRE2Project/pcre2/blob/4a6a8b056f39079d5e958eac84c2ad173f4680bc/ChangeLog#L1050-L1054.

@phil-blain
Copy link
Author

Two things to mention in retrospect:

  1. The commit that added pcre2 support, 8e61838, could have adjusted the help text when compiling with PCRE2 to mention that PCRE regexes are used, and that -E has no effect, since a user in general does not know how the program was compiled.
  2. The changes to the configure script could be improved, since if we pass --with-pcre2=/path/to/self/compiled/pcre2, it results only in -lpcre2posix being added to LIBS, but not in -I/path/to/self/compiled/pcre2/include being added to CFLAGS, nor -L/path/to/self/compiled/pcre2/lib being added to LDFLAGS, so we end up using the system-installed version of the library.

@KangOl CC-ing you since you authored that commit :)

@sergiomb2
Copy link
Contributor

Can you resume the problem ? it is only on RHEL 8 ?

@phil-blain
Copy link
Author

phil-blain commented Mar 7, 2024

Hi @sergiomb2, in the end, the behaviour I was seeing was caused by LD_PRELOAD being set by the sysadmins on the system I was using, which I had forgotten about. This is what I wrote in: #60 (comment). There is nothing much to be done in patchutils on that front, except maybe not using the POSIX names regcomp and regexec when compiled with PCRE2, and instead using pcre2_regcomp and pcre2_regexec. This would have alleviated the problem.

So the issue can be closed, although I noted two possible improvements in #60 (comment).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants