mirror of
https://github.com/JGRennison/OpenTTD-patches.git
synced 2024-11-19 15:25:39 +00:00
872fdcbbab
This allows for v2 container NewGRFs to be created. Hashes are needed for building the baseset metadata, so the .hash files are stored the source tree so they can be built if grfcodec and grfid are not present. The hash of openttd.grf is not actually needed, but it is simpler to leave it in than handle only orig_extra.grf.
65 lines
2.0 KiB
CMake
65 lines
2.0 KiB
CMake
cmake_minimum_required(VERSION 3.5)
|
|
|
|
#
|
|
# Create a single baseset meta file with the correct translations.
|
|
#
|
|
|
|
if(NOT BASESET_SOURCE_FILE)
|
|
message(FATAL_ERROR "Script needs BASESET_SOURCE_FILE defined")
|
|
endif()
|
|
if(NOT BASESET_BINARY_FILE)
|
|
message(FATAL_ERROR "Script needs BASESET_BINARY_FILE defined")
|
|
endif()
|
|
if(NOT BASESET_EXTRAGRF_FILE)
|
|
message(FATAL_ERROR "Script needs BASESET_EXTRAGRF_FILE defined")
|
|
endif()
|
|
|
|
set(ARGC 1)
|
|
set(ARG_READ NO)
|
|
|
|
# Read all the arguments given to CMake; we are looking for -- and everything
|
|
# that follows. Those are our language files.
|
|
while(ARGC LESS CMAKE_ARGC)
|
|
set(ARG ${CMAKE_ARGV${ARGC}})
|
|
|
|
if(ARG_READ)
|
|
list(APPEND LANG_SOURCE_FILES "${ARG}")
|
|
endif()
|
|
|
|
if(ARG STREQUAL "--")
|
|
set(ARG_READ YES)
|
|
endif()
|
|
|
|
math(EXPR ARGC "${ARGC} + 1")
|
|
endwhile()
|
|
|
|
# Place holder format is @<ini_key>_<str_id>@
|
|
file(STRINGS "${BASESET_SOURCE_FILE}" PLACE_HOLDER REGEX "^@")
|
|
string(REGEX REPLACE "@([^_]+).*@" "\\1" INI_KEY "${PLACE_HOLDER}")
|
|
string(REGEX REPLACE "@[^_]+_(.*)@" "\\1" STR_ID "${PLACE_HOLDER}")
|
|
string(REGEX REPLACE "@(.*)@" "\\1" PLACE_HOLDER "${PLACE_HOLDER}")
|
|
|
|
# Get the translations
|
|
foreach(LANGFILE IN LISTS LANG_SOURCE_FILES)
|
|
file(STRINGS "${LANGFILE}" LANGLINES REGEX "^(##isocode|${STR_ID})" ENCODING UTF-8)
|
|
string(FIND "${LANGLINES}" "${STR_ID}" HAS_STR_ID)
|
|
if(HAS_STR_ID LESS 0)
|
|
continue()
|
|
endif()
|
|
string(REGEX REPLACE "##isocode ([^;]+).*" "\\1" ISOCODE "${LANGLINES}")
|
|
if("${ISOCODE}" STREQUAL "en_GB")
|
|
string(REGEX REPLACE "[^:]*:(.*)" "${INI_KEY} = \\1" LANGLINES "${LANGLINES}")
|
|
else()
|
|
string(REGEX REPLACE "[^:]*:(.*)" "${INI_KEY}.${ISOCODE} = \\1" LANGLINES "${LANGLINES}")
|
|
endif()
|
|
list(APPEND ${PLACE_HOLDER} ${LANGLINES})
|
|
endforeach()
|
|
list(SORT ${PLACE_HOLDER})
|
|
string(REPLACE ";" "\n" ${PLACE_HOLDER} "${${PLACE_HOLDER}}")
|
|
|
|
# Get the grf md5
|
|
file(READ ${BASESET_EXTRAGRF_FILE}.hash ORIG_EXTRA_GRF_MD5)
|
|
string(STRIP ${ORIG_EXTRA_GRF_MD5} ORIG_EXTRA_GRF_MD5)
|
|
|
|
configure_file(${BASESET_SOURCE_FILE} ${BASESET_BINARY_FILE})
|