Split aux tag adjustment from read filtering

Pull the part that implements the `--keep-tags` and `--remove-tags`
options out of process_aln() and put it in a new adjust_tags()
function.  This means the tag adjustment only has to be done when
actually needed - so for example it can be skipped in counting
mode.

Invalid aux tags detected in adjust_tags() are now reported
immediately as errors.  This just means they're trapped a bit
earlier, as before they would have caused the writer to fail.

Note that as this code used to come after all the process_aln()
filters had run, the tags were only ever edited if the read
passed filtering.  For compatibility the new version has been
made to work the same way, and notes have been added to the manual
page stating that the options only affect passed reads.

doc/samtools-view.1

@@ -390,12 +390,16 @@ can be repeated multiple times.
 If the list starts with a `^' then it is negated and treated as a
 request to remove all tags except those in \fISTR\fR. The list may be
 empty, so \fB-x ^\fR will remove all tags.
+
+Note that tags will only be removed from reads that pass filtering.
 .TP
 .BI "--keep-tag " STR
 This keeps \fIonly\fR tags listed in \fISTR\fR and is directly equivalent
 to \fB--remove-tag ^\fR\fISTR\fR.  Specifying an empty list will remove
 all tags.  If both \fB--keep-tag\fR and \fB--remove-tag\fR are
 specified then \fB--keep-tag\fR has precedence.
+
+Note that tags will only be removed from reads that pass filtering.
 .TP
 .BR -B ", " --remove-B
 Collapse the backward CIGAR operation.

sam_view.c

@@ -206,6 +206,11 @@ static int process_aln(const sam_hdr_t *h, bam1_t *b, samview_settings_t* settin
         const char *p = bam_get_library((sam_hdr_t*)h, b);
         if (!p || strcmp(p, settings->library) != 0) return 1;
     }
+    return 0;
+}
+
+static int adjust_tags(const sam_hdr_t *h, bam1_t *b,
+                       samview_settings_t* settings) {
     if (settings->keep_tag) {
         uint8_t *s_from, *s_to, *end = b->data + b->l_data;
         auxhash_t h = settings->keep_tag;
@@ -217,7 +222,7 @@ static int process_aln(const sam_hdr_t *h, bam1_t *b, samview_settings_t* settin
             if (s == NULL) {
                 print_error("view", "malformed aux data for record \"%s\"",
                             bam_get_qname(b));
-                break;
+                return -1;
             }
 
             if (kh_get(aux_exists, h, x) != kh_end(h) ) {
@@ -239,7 +244,7 @@ static int process_aln(const sam_hdr_t *h, bam1_t *b, samview_settings_t* settin
             if (s == NULL) {
                 print_error("view", "malformed aux data for record \"%s\"",
                             bam_get_qname(b));
-                break;
+                return -1;
             }
 
             if (kh_get(aux_exists, h, x) == kh_end(h) ) {
@@ -654,8 +659,13 @@ static int fetch_pairs_collect_mates(samview_settings_t *conf, hts_itr_multi_t *
              k = kh_get(names,mate_names,bam_get_qname(rec));
              if ( k != kh_end(mate_names) ) drop = 0;
         }
-        if ( !drop && check_sam_write1(conf->out,conf->header,rec,conf->fn_out,&write_error) < 0 )
-            goto out;
+        if (!drop) {
+            if (adjust_tags(conf->header, rec, conf) != 0)
+                goto out;
+            if (check_sam_write1(conf->out, conf->header, rec, conf->fn_out,
+                                 &write_error) < 0)
+                goto out;
+        }
     }
 
     if (r < -1) {
@@ -685,6 +695,8 @@ static inline int process_one_record(samview_settings_t *conf, bam1_t *b,
     if (!process_aln(conf->header, b, conf)) {
         if (!conf->is_count) {
             change_flag(b, conf);
+            if (adjust_tags(conf->header, b, conf) != 0)
+                return -1;
             if (check_sam_write1(conf->out, conf->header,
                                  b, conf->fn_out, write_error) < 0) {
                 return -1;