also import GDR50 and GDR250
replaces *.shp import with *.gdb import for GDB10.
This commit is contained in:
parent
6f14c17a09
commit
9e7d66e835
44
Makefile
44
Makefile
@ -74,6 +74,9 @@ test-rivers: .faux_test-rivers ## Rivers tests (slow)
|
||||
.PHONY: slides
|
||||
slides: $(SLIDES)
|
||||
|
||||
.PHONY: refresh-rivers
|
||||
refresh-rivers: rivers.sql rivers-50.sql rivers-250.sql ## Refresh river data from national datasets
|
||||
|
||||
###########################
|
||||
# The report, quick version
|
||||
###########################
|
||||
@ -209,9 +212,13 @@ salvis-wm-overlaid-250k-zoom_1SELECT = wm_visuals where name='salvis'
|
||||
salvis-wm-overlaid-250k-zoom_2SELECT = wm_visuals where name='salvis-wm-220'
|
||||
salvis-wm-overlaid-250k-zoom_1COLOR = orange
|
||||
|
||||
.faux_db: db init.sql rivers.sql
|
||||
.faux_db_pre: db init.sql
|
||||
bash db start
|
||||
bash db -f init.sql -f rivers.sql
|
||||
bash db -f init.sql
|
||||
touch $@
|
||||
|
||||
.faux_db: .faux_db_pre rivers.sql
|
||||
bash db -f rivers.sql
|
||||
touch $@
|
||||
|
||||
.faux_test: test.sql wm.sql .faux_db
|
||||
@ -273,8 +280,8 @@ mj-msc-gray.pdf: mj-msc.pdf
|
||||
clean: ## Clean the current working directory
|
||||
-bash db stop
|
||||
-rm -r .faux_test .faux_aggregate-rivers .faux_test-rivers .faux_visuals \
|
||||
.faux_db version.inc.tex vars.inc.tex version.aux version.fdb_latexmk \
|
||||
_minted-mj-msc \
|
||||
.faux_db .faux_db_pre version.inc.tex vars.inc.tex version.aux \
|
||||
version.fdb_latexmk _minted-mj-msc .tmp \
|
||||
$(shell git ls-files -o mj-msc*) \
|
||||
$(addsuffix .pdf,$(FIGURES)) \
|
||||
$(addsuffix .pdf,$(RIVERS)) \
|
||||
@ -287,7 +294,8 @@ clean-tables: ## Remove tables created during unit or rivers tests
|
||||
|
||||
.PHONY: help
|
||||
help: ## Print this help message
|
||||
@awk -F':.*?## ' '/^[a-z0-9.-]*: *.*## */{printf "%-18s %s\n",$$1,$$2}' $(MAKEFILE_LIST)
|
||||
@awk -F':.*?## ' '/^[a-z0-9.-]*: *.*## */{printf "%-18s %s\n",$$1,$$2}' \
|
||||
$(MAKEFILE_LIST)
|
||||
|
||||
.PHONY: wc
|
||||
wc: mj-msc.pdf ## Character and page count
|
||||
@ -296,21 +304,15 @@ wc: mj-msc.pdf ## Character and page count
|
||||
tr -d '[:space:]' | wc -c | \
|
||||
awk '{printf("Chars: %d, pages: %.1f\n", $$1, $$1/1500)}'
|
||||
|
||||
.PHONY: refresh-rivers
|
||||
refresh-rivers: aggregate-rivers.sql .faux_db ## Refresh rivers.sql from GDB10LT
|
||||
@if [ ! -f "$(GDB10LT)" ]; then \
|
||||
echo "ERROR: GDB10LT-static-*.zip not found. Do GDB10LT=<...>"; \
|
||||
define rivers_template
|
||||
$(1): aggregate-rivers.sql gdr2pgsql .faux_db_pre
|
||||
@if [ ! -f "$$($(2))" ]; then \
|
||||
echo "ERROR: $(2)-static-*.zip not found. Run env $(2)=<...>"; \
|
||||
exit 1; \
|
||||
fi
|
||||
mkdir -p .tmp/shp; unzip -d .tmp/shp "$(GDB10LT)" 'HIDRO_L.*'
|
||||
shp2pgsql -s 3857 -d ".tmp/shp/HIDRO_L.shp" | \
|
||||
awk '!/^INSERT/{print}; /^INSERT/&&/$(RIVERFILTER)/{print;next}' | \
|
||||
bash ./db
|
||||
bash db -f $<
|
||||
(\
|
||||
echo '-- Generated at $(shell TZ=UTC date +"%FT%TZ") on $(shell whoami)@$(shell hostname -f)'; \
|
||||
echo '-- Rivers: $(RIVERFILTER)'; \
|
||||
docker exec wm-mj pg_dump --clean -Uosm osm -t wm_rivers | tr -d '\r' \
|
||||
) > rivers.sql.tmp
|
||||
mv rivers.sql.tmp rivers.sql
|
||||
rm -fr .tmp/shp
|
||||
./gdr2pgsql "$$($(2))" "$(3)" "$(RIVERFILTER)" "$(1)"
|
||||
endef
|
||||
|
||||
$(eval $(call rivers_template,rivers.sql,GDB10LT,wm_rivers))
|
||||
$(eval $(call rivers_template,rivers-50.sql,GDR50LT,wm_rivers_50))
|
||||
$(eval $(call rivers_template,rivers-250.sql,GDR250LT,wm_rivers_250))
|
||||
|
@ -1,7 +1,7 @@
|
||||
/* Aggregates rivers by name and proximity. */
|
||||
drop function if exists aggregate_rivers;
|
||||
create function aggregate_rivers() returns table(
|
||||
id bigint,
|
||||
id integer,
|
||||
name text,
|
||||
way geometry
|
||||
) as $$
|
||||
@ -35,15 +35,15 @@ $$ language plpgsql;
|
||||
drop index if exists wm_rivers_tmp_id;
|
||||
drop index if exists wm_rivers_tmp_gix;
|
||||
drop table if exists wm_rivers_tmp;
|
||||
create temporary table wm_rivers_tmp (id bigint, name text, way geometry);
|
||||
create temporary table wm_rivers_tmp (id serial, name text, way geometry);
|
||||
create index wm_rivers_tmp_id on wm_rivers_tmp(id);
|
||||
create index wm_rivers_tmp_gix on wm_rivers_tmp using gist(way) include(name);
|
||||
|
||||
insert into wm_rivers_tmp
|
||||
select p.gid as id, p.vardas as name, p.geom as way from hidro_l p;
|
||||
insert into wm_rivers_tmp (name, way)
|
||||
select p.vardas as name, p.shape as way from :srctable p;
|
||||
|
||||
drop table if exists wm_rivers;
|
||||
create table wm_rivers as (
|
||||
drop table if exists :dsttable;
|
||||
create table :dsttable as (
|
||||
select * from aggregate_rivers() where st_length(way) >= 50000
|
||||
);
|
||||
drop table wm_rivers_tmp;
|
||||
|
34
gdr2pgsql
Executable file
34
gdr2pgsql
Executable file
@ -0,0 +1,34 @@
|
||||
#!/bin/bash
|
||||
set -euo pipefail
|
||||
|
||||
dbzip=$(realpath -s "$1")
|
||||
tbl=$2
|
||||
filter=$3
|
||||
outfile=$4
|
||||
|
||||
if [[ "$dbzip" =~ " " ]]; then
|
||||
echo "ERROR: $dbzip contains spaces"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
set -x
|
||||
|
||||
gdbname=$(unzip -Z1 "$dbzip" | awk -Fgdb '/.gdb/{print $1"gdb";exit}')
|
||||
|
||||
now=$(TZ=UTC date +"%FT%TZ")
|
||||
here=$(whoami)@$(hostname -f)
|
||||
|
||||
mkdir -p ".tmp"
|
||||
ogr2ogr -skipfailures -f PGDump /dev/stdout -t_srs epsg:3857 \
|
||||
"/vsizip/$dbzip/$gdbname" -nln "src_$tbl" hidro_l | \
|
||||
awk "!/^INSERT/{print}; /^INSERT/&&/${filter}/{print;next}" | \
|
||||
bash ./db | \
|
||||
grep -v '^INSERT 0 1'
|
||||
|
||||
bash db -f aggregate-rivers.sql -v "srctable=src_$tbl" -v "dsttable=$tbl"
|
||||
(
|
||||
echo "-- Generated at $now on $here";
|
||||
echo "-- Rivers: $filter";
|
||||
docker exec wm-mj pg_dump --clean -Uosm osm -t "$tbl" | tr -d '\r'
|
||||
) > ".tmp/$outfile"
|
||||
mv ".tmp/$outfile" "$outfile"
|
Loading…
Reference in New Issue
Block a user