[flang][driver] Randomise the names of the temporary unparsed files

[awarzynski]

Summary

This patch makes sure that the base name of the temporary unparsed files
(generated by the flang bash script) are randomised. This is required
to make the name of these temporary files unique to a particular
invocation of the script. Otherwise, we cannot reliably run it in
parallel.

[DavidSpickett]

flang/tools/f18/flang.in

331

Wouldn't date/time have a small chance to collide?

Would something like /proc/sys/kernel/random/uuid be better to guarantee that each one is unique?

[awarzynski]

flang/tools/f18/flang.in

331

Wouldn't date/time have a small chance to collide?

I'm also slightly concerned, but %N gives you nanoseconds and I'd hope that that's enough to get rando results.

Would something like /proc/sys/kernel/random/uuid be better to guarantee that each one is unique?

I like your suggestion! Are we confident that it's going to be present on every system that this script might be run on? I think that it would be OK to limit it to Linux, but I would need to check with @sscalpone (he is likely to use this script a lot).

[tschuett]

flang/tools/f18/flang.in

331

How about mktemp? At least on Apple proc does not exists. I doubt Windows will have a proc dir.

[awarzynski]

flang/tools/f18/flang.in

331

This one is also nice @tschuett , thanks! However, the version that I have on Darwin only allows me to specify the prefix. I need to be able to specify the suffix though:

mktemp XXXXXX.f90

I can do this with mktemp on Linux.

Perhaps it's worth adding a bit of context here. This script is only a temporary workaround and hence there is no need to support every possible system or to make it future-proof. It is fine to use something that's OS specific, as long as we support all our users that rely on this script. AFAIK, the list is not long. And I assume that all of Linaro's BuildBots are running Linux.

As for /proc/sys/kernel/random/uuid, I just want to make sure that it's available on all Linux systems. Given the location, it's seems safe to assume that that's the case, right?

[DavidSpickett]

flang/tools/f18/flang.in

331

I see some isolated comments that it's not present on some Linux, but it's listed on the man page:
https://man7.org/linux/man-pages/man4/random.4.html
(maybe they were talking about older kernels)

I have uuidgen on Ubuntu, maybe that's a better way to go if Macs also have that. (since they and the BSDs won't have /proc)

[DavidSpickett]

flang/tools/f18/flang.in

331

And I assume that all of Linaro's BuildBots are running Linux.

All recent Ubuntu yes.

[awarzynski]

flang/tools/f18/flang.in

331

I have uuidgen on Ubuntu, maybe that's a better way to go if Macs also have that. (since they and the BSDs won't have /proc)

Yes, it's available on Mac OS and it looks like the most adequate solution here. Thanks for the suggestion!

[awarzynski]

Switch from using $RANDOM to uuidgen.

[DavidSpickett]

LGTM

flang/tools/f18/flang.in

main() {
   local flang_options=("-fc1")
   # Options for the external Fortran Compiler
   local ext_fc_options=()
   categorise_opts OPTIONS flang_options ext_fc_options
 
   local -r wd=$(cd "$(dirname "$0")/.." && pwd)
 
   # STEP 1: Unparse
-  local -r unparsed_file="flang_unparsed_source_file"
+  # Base-name for the unparsed files. These are just temporary files that are
+  # first generated and then deleted by this script.
+  # NOTE: We need to make sure that the base-name is unique to every
+  # invocation. Otherwise we can't use this script in parallel.
+  local -r unique_id=$(uuidgen | cut -b25-36)
+  local -r unparsed_file_base="flang_unparsed_file_$unique_id"
+
   flang_options+=("-module-suffix")
   flang_options+=(".f18.mod")
   flang_options+=("-fdebug-unparse")
   flang_options+=("-fno-analyzed-objects-for-unparse")
 
   [[ ! -z ${MODULE_DIR} ]] && flang_options+=("-module-dir ${MODULE_DIR}")
   [[ ! -z ${INTRINSICS_MOD_DIR} ]] && flang_options+=("-intrinsics-module-directory ${INTRINSICS_MOD_DIR}")
   for idx in "${!fortran_source_files[@]}"; do
-    if ! "$wd/bin/@FLANG_DEFAULT_DRIVER@" "${flang_options[@]}" "${fortran_source_files[$idx]}" -o "${unparsed_file}_${idx}.f90"
+    if ! "$wd/bin/@FLANG_DEFAULT_DRIVER@" "${flang_options[@]}" "${fortran_source_files[$idx]}" -o "${unparsed_file_base}_${idx}.f90"
     then status=$?
          echo flang: in "$PWD", @FLANG_DEFAULT_DRIVER@ failed with exit status $status: "$wd/bin/@FLANG_DEFAULT_DRIVER@" "${flang_options[@]}" "$@" >&2
          exit $status
     fi
   done
 
   # STEP 2: Compile Fortran Source Files
   readonly ext_fc="${F18_FC:-gfortran}"
   for idx in "${!fortran_source_files[@]}"; do
     # We always have to specify the output name with `-o <out_obj_file>`. This
     # is because we are using the unparsed rather than the original source file
     # below. As a result, we cannot rely on the compiler-generated output name.
     out_obj_file=$(get_relocatable_name "${fortran_source_files[$idx]}")
 
-    if ! $ext_fc "-c" "${ext_fc_options[@]}" "${unparsed_file}_${idx}.f90" "-o" "${out_obj_file}"
+    if ! $ext_fc "-c" "${ext_fc_options[@]}" "${unparsed_file_base}_${idx}.f90" "-o" "${out_obj_file}"
     then status=$?
       echo flang: in "$PWD", "$ext_fc" failed with exit status $status: "$ext_fc" "${ext_fc_options[@]}" "$@" >&2
          exit $status
     fi
     object_files+=(${out_obj_file})
   done
 
   # Delete the unparsed files
   for idx in "${!fortran_source_files[@]}"; do
-    rm "${unparsed_file}_${idx}.f90"
+    rm "${unparsed_file_base}_${idx}.f90"
   done
 
   # STEP 3: Compile Other Source Files
   for idx in "${!other_source_files[@]}"; do
     # We always specify the output name with `-o <out_obj_file>`. The user
     # might have used `-o`, but we never add it to $OPTIONS (or
     # $ext_fc_options). Hence we need to use `get_relocatable_name`.
     out_obj_file=$(get_relocatable_name "${other_source_files[$idx]}")