source: BLFS/libs/func_dependencies@ 88e5cbc

ablfs-more trunk
Last change on this file since 88e5cbc was abadee3, checked in by Pierre Labastie <pierre.labastie@…>, 3 years ago

Improve and augment the "func_dependency" doc

  • Property mode set to 100644
File size: 27.9 KB
Line 
1#!/bin/bash
2#
3
4#-----------------------------------------------------------------------------#
5# This is a set of (recursive) functions for manipulating a dependency graph. #
6# We use algorithms and definitions from chapter 4 (mainly section 4.2) of #
7# https://algs4.cs.princeton.edu/. The graph we manipulate is the directed #
8# graph of the dependencies: nodes are packages in the BLFS book. A node A is #
9# connected to a node B if package A depends on B. A topological order (rev- #
10# erted) is exactly what we want for a build order. But a topological order #
11# only exists if the graph is acyclic. We'll therefore have to remove cycles. #
12# There are a number of other features we want to consider: #
13# - edges are weighted according to the dependency requirement: #
14# 1 for required #
15# 2 for recommended #
16# 3 for optional #
17# 4 for external #
18# We should consider only edges with weight lower or equal to that #
19# specified by the user, but see below. #
20# - we do not want to build the whole book. The user requests a set of #
21# packages, and we'd like to consider only nodes reachable from this set #
22# using edges of weight not exceeding the specified weight. #
23# - we do not want to rebuild packages already built. But we still have to #
24# generate the full dependency graph, because if A depends on B, which is #
25# already built, and B depends on C, which is not built, or needs to be #
26# updated, then A may depends on C. We therefore have to remove already #
27# built (and up to date) packages from the graph, but need to keep the #
28# dependency chain. #
29# - when doing the topological sort, we want to consider all the edges and #
30# not only those not exceeding the specified weight: If a package A in the #
31# reachable subgraph depends optionally on another package B in the same #
32# subgraph, we want to build B before A if possible. But this means we'll #
33# have to remove cycles for all weights. #
34# - dependencies have another qualifier: before, after, or first. We use #
35# it as follows: for "after", we can build the dependency after the #
36# package, but if a package A depends on B with an "after" qualifier, and #
37# a package C depends on A with a "before" qualifier, C may need B to be #
38# able to use A. So the only safe way to consider "after" qualifiers is to #
39# consider that they are "before" deps for any parent of the packages #
40# considered. There is an exception to that rule: if B depends on C #
41# (possibly through a chain of several dependencies), then C should still #
42# be built before B. For "after", the dependency has to be built both #
43# before and after the package. So we duplicate the dependency as a #
44# "-pass1" package, and change the graph accordingly. #
45# We'll therefore have a 3 pass procedure. First build the set of nodes #
46# reachable from the root set. Second, remove dangling edges (those pointing #
47# to packages outside the node set), and move "after" edges to "before" edges #
48# originating from the parents as well as creating the "-pass1" nodes. Third #
49# remove cycles and generate a topological sort. #
50# #
51# Pass 1: graph generation #
52# ======================== #
53# Data layout for pass 1 #
54# ---------------------- #
55# A node of the graph is represented by a text file <nodeName>.dep. Each edge #
56# starting from this node is represented by a line in this file. We keep #
57# those files in the same directory. We introduce a special node named root, #
58# whose edges point to the list of nodes requested by the user. Each line #
59# contains three fields: #
60# - the weight of the edge #
61# - the qualifier: "before" (b), "after" (a), or "first" (f) #
62# - the name of the destination of the edge (without the ".dep" extension) #
63# #
64# Recursive function "generate_subgraph" #
65# -------------------------------------- #
66# This function treats a node of the graph that is not a leaf and that is #
67# seen for the first time in the DFS. The dependencies of this node are #
68# known, and stored in a .dep file. For each dependency in that file, there #
69# are three cases: #
70# - the weight of the edge leading to that dependency is higher than #
71# requested. This dependency is discarded (some information printed) #
72# - the weight of the edge is lower or equal to requested, but the node #
73# has already been visited (the .dep file exists). Discard too (some #
74# information printed) #
75# - the weight of the edge is lower or equal to requested, and the node #
76# has not been seen: then the dependencies of that node are generated, #
77# and there are two cases: #
78# - the node has no dependencies: just create an empty .dep file, so #
79# that we know the node has been visited #
80# - the node has dependencies: call generate_subgraph for that node #
81# #
82# This function takes four parameters: #
83# - The node filename: this is the only one useful for the algorithm #
84# - The depth: number of steps starting from root (for pretty print only) #
85# - The weight of the edge leading to that node (for printing) #
86# - The qualifier (for printing) #
87# #
88# Pass 2: graph transformation #
89# ============================ #
90# We now have three loops over nodes of the graph #
91# Loop 1: Remove dead edges #
92# ------------------------- #
93# Since some nodes have not been created because the edges leading to them #
94# had too high a weight, those edges have to be suppressed. #
95# For each existing node file, we make a list of lines to remove by #
96# testing whether the destination exists. We then remove the lines. #
97# Another approach would be to make a temporary file and output only #
98# lines that should stay, then rename the file. This would save a loop. #
99# All in all it is an N*e process, where N is the number of nodes and e #
100# the average number of edges originating from a node. #
101# Loop 2: Treat "after" edges #
102# --------------------------- #
103# If a node is the origin of edges qualified as "after", we want the #
104# nodes which are the destination of those edges to be built after #
105# the origin node, but before any node that depend on the origin #
106# node. For that, the general rule is to change: #
107# P---b--->A---a--->D #
108# to: #
109# P---b--->Agroupxx---b--->A #
110# | #
111# ---b--->D #
112# But there is a problem if D depends on P, possibly through a chain, #
113# because we create a cycle which shouldn't exist. If this is the case, #
114# we leave A as a dependency of P: #
115# P---b--->A #
116# #
117# Agroupxx---b--->A #
118# | #
119# ---b--->D #
120# Doing so, it may happen that Agroupxx has no parent. We then add #
121# Agroupxx as a dependency of root. The problem with this algorithm is #
122# the search for paths from D to A, which may be exponential in the #
123# number of nodes in the graph. #
124# #
125# Loop 3: Add -pass1 nodes #
126# ------------------------ #
127# Sometimes there is no way to escape a cycle. A package A needs B, and B #
128# needs A. In that case, it is often possible to build a degraded version #
129# of package A, then B, then rebuild A. The book indicates this with the #
130# following dependency chain, using a qualifier of "first": #
131# B---f--->A---b--->X...Y---b--->B #
132# where the X...Y notation represents a chain of dependencies from A to B. #
133# So the third loop is over nodes containing "f" qualifiers, and does the #
134# following: it creates a new node A-pass1, which is a copy of A, and #
135# remove from A-pass1 all the dependencies leading to B through a chain, #
136# to obtain: #
137# A---b--->X...Y---b--->B---b--->A-pass1 #
138# It may then happen that nothing depends on A. So this is tested, and A #
139# is added to the root node if it is orphaned. #
140# TODO: document the third pass #
141# TODO: needs also to document the .tree files #
142# TODO: The following is obsolete #
143# Circular dependencies: #
144# #
145# In case we find a cirdular dependency, it has the form : #
146# parent->dependency_0->...->dependency_n->dependency_0 #
147# If we want to build dependency_n before dependency_0, no problem: #
148# we just prune the tree at dependency_n. If we want to build first #
149# dependency_0, we need to put dependency_n as a dependency of parent, #
150# then erase and rebuild the subtree from there. Now, we may have met #
151# another circular dependency in the subtree, and erasing the tree makes #
152# us forget the decision which was made. So, after first generating the #
153# list of dependencies from packages.xml, we keep the generated list in #
154# a file <nodeName>.odep, which we modify according to the decision which #
155# was made. #
156#---------------------------------------------------------------------------#
157
158# Global variables:
159# A string of spaces for indenting:
160declare -a spaceSTR=" "
161# When we are backing up from a circular dependency, `parentNode'
162# contains the node which has an edge entering the cycle
163declare parentNode
164
165#---------------------#
166generate_subgraph() { #
167#---------------------#
168: <<inline_doc
169 function: Create a subgraph of all the nodes reachable from the node
170 represented by the file whose name is $1. The edges considered
171 are those with maximal weight DEP_LEVEL (recursive function).
172 input vars: $1 : file name corresponding to the node whose edges will be
173 followed for the DFS
174 $2 : weight of the edge leading to this node
175 $3 : depth (root is 1)
176 $4 : qualifier (a for after, b for before, f for first)
177 externals: vars: DEP_LEVEL contains 1 if we want to build the
178 tree only for required dependencies,
179 2 if we want also recommended ones,
180 3 if we want also optional ones, but only
181 for the requested packages,
182 4 if we want all the dependencies
183 (excluding external of course)
184 MAIL_SERVER contains the name of the MTA we want to use.
185 files: ../xsl/dependencies.xsl: stylesheet for creating the
186 .dep files
187 ../packages.xml: File containing packages id
188 and dependencies
189 returns: 0 if the tree has been successfully created
190 output: files: for each node reachable from $1, a file <node>.dep.
191 on error: nothing
192 on success: nothing
193inline_doc
194
195local depFile=$1
196local -i weight=$2
197local -i depth=$3
198local qualifier=$4
199local -i spacing=0
200local priostring
201local buildstring
202local id_of_dep
203local prio_of_dep
204local build_of_dep
205local dep_level
206
207if (( depth < 10 )); then spacing=1; fi
208case $weight in
209 1) priostring=required ;;
210 2) priostring=recommended ;;
211 3) priostring=optional ;;
212esac
213case $qualifier in
214 a) buildstring=runtime ;;
215 b|f) buildstring= ;;
216esac
217dep_level=$DEP_LEVEL
218if [ "$dep_level" = 3 ] && [ "$depth" -gt 2 ]; then dep_level=2; fi
219if [ "$dep_level" -gt 3 ]; then dep_level=3; fi
220echo -en "\nNode: $depth${spaceSTR:0:$(( depth + spacing ))}${RED}${depFile%.dep}${OFF} $priostring $buildstring"
221
222depth=$(( depth + 1 ))
223if (( depth < 10 )); then spacing=1; else spacing=0; fi
224# Start of loop
225{
226while read prio_of_dep build_of_dep id_of_dep; do
227 case $prio_of_dep in
228 1) priostring=required ;;
229 2) priostring=recommended ;;
230 3) priostring=optional ;;
231 4) priostring=external ;;
232 esac
233 case $build_of_dep in
234 a ) buildstring=runtime ;;
235 b|f) buildstring= ;;
236 esac
237# Has this entry already been seen?
238# TODO: no there is no special case!
239# We have a special case here: if the entry has been seen at depth > 2
240# and now depth=2 and DEP_LEVEL=3, optional deps have not been processed.
241# If this is the case, just consider it has not been seen.
242 if [ -f ${id_of_dep}.dep ] ; then
243 case $depth$DEP_LEVEL in
244 23) ;;
245 *)
246# Just display it and proceed.
247 echo -en "\nEdge: $depth${spaceSTR:0:$((depth + spacing))}${MAGENTA}${id_of_dep}${OFF} $priostring $buildstring"
248 continue
249 ;;
250 esac
251 fi
252# Is the weight higher than requested?
253 if [ "$prio_of_dep" -gt $dep_level ]; then
254# Just display it and proceed.
255 echo -en "\n Out: $depth${spaceSTR:0:$((depth + spacing))}${YELLOW}${id_of_dep}${OFF} $priostring $buildstring"
256 continue
257 fi
258# Otherwise, let's build the corresponding subgraph.
259 xsltproc --stringparam idofdep "$id_of_dep" \
260 --stringparam MTA "$MAIL_SERVER" \
261 -o ${id_of_dep}.dep \
262 ../xsl/dependencies.xsl ../packages.xml
263
264 if [[ -s ${id_of_dep}.dep ]]; then # this dependency has dependencies
265 generate_subgraph ${id_of_dep}.dep $prio_of_dep $depth $build_of_dep
266 else # id_of_dep has no dependencies, just touch the file and display
267 touch ${id_of_dep}.dep
268 echo -en "\nLeaf: $depth${spaceSTR:0:$((depth + spacing))}${CYAN}${id_of_dep}${OFF} $priostring $buildstring"
269 fi
270done
271} <$depFile
272depth=$(( depth - 1 ))
273if (( depth < 10 )); then spacing=1; else spacing=0; fi
274echo -en "\n End: $depth${spaceSTR:0:$((depth + spacing))}${GREEN}${depFile%.dep}${OFF}"
275return 0
276}
277
278#-----------#
279path_to() { #
280#-----------#
281: <<inline_doc
282 function: check whether there is a path from $1 to $2 on the graph
283 input vars: $1 contains the filename of the starting node.
284 $2 contains the name of the node to reach
285 $3 contains the weight above which we do not want to
286 follow an edge
287 $seen (global) contains the list of already seen nodes.
288 It must ve set to " " prior to calling the function
289 returns: 0 if the node has been found
290 1 if not
291 on error: nothing
292 on success: nothing
293inline_doc
294local start=$1
295local seek=$2
296local prio=$3
297local prio_of_dep
298local build_of_dep
299local id_of_dep
300local r
301
302if test "${start%.dep}" = "$seek"; then return 0; fi
303seen="$seen${start%.dep} "
304if test -s $start; then
305 {
306 while read prio_of_dep build_of_dep id_of_dep; do
307 if test "$prio" -lt "$prio_of_dep"; then continue; fi
308 if ! test "${seen% $id_of_dep *}" = "$seen"; then continue; fi
309 if path_to ${id_of_dep}.dep $seek $prio; then return 0; fi
310 done
311 } < $start
312fi
313return 1
314}
315#------------------#
316clean_subgraph() { #
317#------------------#
318: <<inline_doc
319 function: Remove dangling edges and create groups of deps for "after"
320 deps: A-before->B-after->C becomes:
321 A -before-> Bgroupxx -before-> B
322 \
323 -before-> C
324 the name of the group is chosen so that it is unlikely as
325 a package name (so that it is removed when building the
326 xml book).
327 Also change the "first" qualifier so that a cycle:
328 A -first-> B ---chain---> A becomes:
329 B ---chain---> A -before-> B-pass1
330 and we remove all the dependencies which have a chain to
331 A in B-pass1.
332 Since we do not change anything else, it may happen that
333 nothing depends on B. In that case, B is appended to root.
334 input vars: None
335 files: <node>.dep files containing dangling edges and
336 "after" qualifiers
337 returns: 0
338 output: files: <node>.dep files containing no dangling edges and
339 no "after" qualifiers
340 on error: nothing
341 on success: nothing
342inline_doc
343
344local node
345local id_of_dep
346local prio_of_dep
347local build_of_dep
348local lines_to_remove
349local lines_to_change
350local parent
351local p
352local b
353local start
354local seen
355
356for node in $(ls *.dep); do
357 if test $node = root.dep; then continue; fi
358 echo Cleaning $node
359 lines_to_remove=
360 {
361 while read prio_of_dep build_of_dep id_of_dep; do
362 if ! test -f ${id_of_dep}.dep; then
363 lines_to_remove="$lines_to_remove $id_of_dep"
364 continue
365 fi
366 done
367 } <$node
368 for id_of_dep in $lines_to_remove; do
369 sed "/\ $id_of_dep\$/d" -i $node
370 done
371done
372for node in $(grep -l ' a ' *.dep); do
373 lines_to_remove=
374 echo Process "runtime" deps in $node
375 if ! [ -e ${node%.dep}groupxx.dep ]; then
376 b=0 # Nothing depends on <node>groupxx
377 for parent in $(grep -l ${node%.dep}\$ *); do
378 p=0 # No "after" dependency depends on this parent
379 for start in $(grep ' a ' $node | cut -d' ' -f3); do
380 seen=" " # global variable used in "path_to"
381 if path_to ${start}.dep ${parent%.dep} 3; then p=1; break; fi
382 done
383 if test $p = 0; then
384 b=1
385 sed -i "s/\ ${node%.dep}\$/&groupxx/" $parent
386 fi
387 done
388 echo "1 b ${node%.dep}" > ${node%.dep}groupxx.dep
389 if test $b = 0; then echo "1 b ${node%.dep}groupxx" >> root.dep; fi
390 fi
391 {
392 while read prio_of_dep build_of_dep id_of_dep; do
393 if test $build_of_dep = a; then
394 if ! grep -q ${id_of_dep} ${node%.dep}groupxx.dep; then
395 echo "$prio_of_dep b ${id_of_dep}" >> ${node%.dep}groupxx.dep
396 fi
397 lines_to_remove="$lines_to_remove $id_of_dep"
398 fi
399 done
400 } <$node
401 for id_of_dep in $lines_to_remove; do
402 sed "/\ $id_of_dep\$/d" -i $node
403 done
404done
405for node in $(grep -l ' f ' *); do
406 echo Process "first" deps in $node
407 lines_to_change=
408 {
409 while read prio_of_dep build_of_dep id_of_dep; do
410 if test $build_of_dep = f; then
411 if ! test -f ${id_of_dep}-pass1.dep; then
412 cp ${id_of_dep}{,-pass1}.dep;
413 fi
414 lines_to_change="$lines_to_change $id_of_dep"
415 unset lr # lines to remove in -pass1
416 {
417 while read p b start; do
418 seen=" " # global variable used in "path_to"
419 if path_to ${start}.dep ${node%.dep} $p; then
420 lr="$lr $start"
421 fi
422 done
423 } < ${id_of_dep}-pass1.dep
424 for p in $lr; do
425 sed "/\ $p\$/d" -i ${id_of_dep}-pass1.dep
426 done
427 fi
428 done
429 } <$node
430 for id_of_dep in $lines_to_change; do
431 sed "/\ $id_of_dep\$/"'{s/[[:digit:]] f /1 b /;s/$/-pass1/}' -i $node
432 if ! grep -q " $id_of_dep\$" *.dep; then
433 echo 1 b $id_of_dep >>root.dep
434 fi
435 done
436done
437} # End clean_subgraph
438
439#----------------------------#
440generate_dependency_tree() { #
441#----------------------------#
442: <<inline_doc
443 function: Create a subtree of the dependency tree
444 (recursive function)
445 input vars: $1 : file with a list of targets (child nodes)
446 the first line of the file is the link
447 $2 : priority (1=req, 2=rec, 3=opt)
448 returns: 0 if the tree has been successfully created
449 1 if we are backing up to the parent of a circular dep
450 and there are only required deps in the cycle
451 2 if we are backing up to the parent of a circular dep
452 and there are recommended deps and no optional deps in the
453 cycle
454 3 if we are backing up to the parent of a circular dep
455 and there are optiional deps in the cycle
456 modifies: vars: ParentNode is set when return is not 0
457 output: files: for each <pkg> with dependencies in $1,
458 a file <pkg>.tree and its dependencies
459 on error: nothing
460 on success: nothing
461inline_doc
462
463local depFile=$1
464local priority=$2
465local -a rootlink
466local -a priolink
467local -a otherlink
468local -i depth
469local -i count=0
470local id_of_dep
471local build_of_dep
472local prio_of_dep
473local parent
474local lines_to_remove=
475local srootlink
476local priostring
477local dpriostring
478local i
479
480{
481read -a rootlink
482depth=${#rootlink[*]}
483read -a priolink
484srootlink="${rootlink[*]} "
485case $priority in
486 1) priostring=required ;;
487 2) priostring=recommended ;;
488 3) priostring=optional ;;
489esac
490# start of depFile
491echo -en "\nNode: $depth${spaceSTR:0:$depth}${RED}${depFile%.tree}${OFF} $priostring"
492
493while read prio_of_dep build_of_dep id_of_dep; do
494 case $prio_of_dep in
495 1) dpriostring=required ;;
496 2) dpriostring=recommended ;;
497 3) dpriostring=optional ;;
498 esac
499# count entries in file
500 (( count++ ))
501# Has this entry already been seen?
502 if [ -f ${id_of_dep}.tree ]; then # found ${id_of_dep}.tree already in tree
503 otherlink=($(head -n 1 ${id_of_dep}.tree))
504 if [ -z "${otherlink[*]}" ]
505 then echo otherlink empty for $id_of_dep.tree
506 echo This should not happen, but happens to happen...
507 exit 1
508 fi
509#Do not use "${rootlink[*]}" =~ "${otherlink[*]}": case rootlink=(1 11)
510# and otherlink=(1 1)
511 if [[ ${srootlink#"${otherlink[*]} "} != ${srootlink} ]]; then # cir. dep
512 echo -en "\nCirc: $((depth+1))${spaceSTR:0:$((depth+1))}${YELLOW}${id_of_dep}${OFF} $dpriostring"
513# Find lowest priority in branch from parent to depFile:
514 p2=0
515 for (( i=${#otherlink[*]}; i < $depth ; i++ )) ; do
516 if (( ${priolink[i]} > $p2 )); then p2=${priolink[i]}; fi
517 done
518 if (( $prio_of_dep >= $p2 )); then # prune
519 lines_to_remove="$lines_to_remove $id_of_dep"
520 sed -i "/$id_of_dep/d" ${depFile/.tree/.dep}
521 else # find and set parent, then return lowest priority
522# The parent has the same link without the last entry.
523# We do not need otherlink anymore so just destroy the last element
524 unset otherlink[-1]
525 parentNode=$(grep ^"${otherlink[*]}"\$ -l *)
526 return $p2
527 fi
528 else # not circular: prune tree (but not .dep, since it may happen that
529 # the tree is destroyed and rebuilt in another order)
530 lines_to_remove="$lines_to_remove $id_of_dep"
531 fi # circular or not
532 continue # this dependency has already been seen, and the associated
533 # subtree computed. We are done
534 fi # Has this entry already been seen?
535# So, this entry has not already been seen. Let's build the corresponding
536# subtree. First check there is a subtree...
537# Use -s, because it may happen that after removing lines, .dep exists
538# but is empty.
539 if [[ -s ${id_of_dep}.dep ]]; then # this dependency has dependencies
540 sed "1i${rootlink[*]} $count\\
541${priolink[*]} $prio_of_dep" < ${id_of_dep}.dep \
542 > ${id_of_dep}.tree # add link and priolink
543 generate_dependency_tree ${id_of_dep}.tree $prio_of_dep
544# Test return value, in case we exchange dependencies
545 p2=$?
546 case $p2 in
547 0) # Normal return
548 ;;
549 $prio_of_dep) # we remove this dep, but since it may become unreachable,
550 # move it to be built later (as a dep of parent).
551 tree_erase ${id_of_dep}.tree
552 lines_to_remove="$lines_to_remove $id_of_dep"
553 sed -i "/${id_of_dep}/d" ${depFile/.tree/.dep}
554 echo "$prio_of_dep b $id_of_dep" >> $parentNode
555# must be added to .dep in case parentNode is destroyed when erasing
556# the tree
557 echo "$prio_of_dep b $id_of_dep" >> ${parentNode/.tree/.dep}
558 continue
559 ;;
560 *) # We are backing up
561 return $p2
562 ;;
563 esac
564 else # id_of_dep has no dependencies, just record the link in a file
565 # and print
566 echo "${rootlink[*]} $count" > ${id_of_dep}.tree
567 echo -en "\nLeaf: $(($depth+1))${spaceSTR:0:$(($depth+1))}${CYAN}${id_of_dep}${OFF} $dpriostring"
568 fi
569done
570echo -en "\n End: $depth${spaceSTR:0:$depth}${GREEN}${depFile%.tree}${OFF}"
571} <$depFile
572# It may happen that a file is created with several times
573# the same line. Normally, all those lines but one
574# would be flagged to be removed (or all of them if
575# the dependency appeared before). A simple sed /$line/d
576# destroys all the lines. We should instead remove
577# only one for each appearance of it in lines_to_remove.
578# so first get the position of last line and then delete
579# that line
580for line in $lines_to_remove
581 do lineno=$(sed -n /^[[:digit:]]\ b\ $line\$/= $depFile | tail -n1)
582 sed -i ${lineno}d $depFile
583done
584return 0
585}
586
587
588#---------------#
589tree_browse() { #
590#---------------#
591local file=$1
592local f
593
594#echo file=$file
595for f in $(grep '[^0-9 ]' $file | sed 's/.* //'); do
596# echo f=$f
597 if grep -q '[^0-9 ]' ${f}.tree ; then
598 tree_browse ${f}.tree
599 fi
600 echo $f
601done
602}
603
604#--------------#
605tree_erase() { #
606#--------------#
607local file=$1
608local f
609local rootlink
610local rootlink2
611
612#echo file=$file
613rootlink="$(head -n1 $file) "
614for f in $(grep '[^0-9 ]' $file | sed 's/.* //'); do
615 if [ -f ${f}.tree ]; then
616 rootlink2="$(head -n1 ${f}.tree) "
617# We want two things:
618# i) do not erase the file if it is in another branch
619# ii) do not erase the file if there is a circular dependency
620# for case i), we test that rootlink is contained in rootlink2
621# for case ii), we test that rootlink2 is not contained in
622# rootlink.
623# See comment above about srootlink
624 if [[ ${rootlink2#${rootlink}} != ${rootlink2} &&
625 ${rootlink#${rootlink2}} == ${rootlink} ]] ; then
626 tree_erase ${f}.tree
627 fi
628 fi
629done
630rm -f $file
631}
Note: See TracBrowser for help on using the repository browser.