libsolv/0005-Use-less-memory-when-extending-packages.patch

266 lines
7.8 KiB
Diff
Raw Normal View History

2016-03-08 12:12:00 +00:00
From d06562f7b70f6ceb7d52e717efd1963ce6e8ecf0 Mon Sep 17 00:00:00 2001
From: Michael Schroeder <mls@suse.de>
Date: Mon, 29 Feb 2016 13:49:48 +0100
Subject: [PATCH 4/9] Use less memory when extending packages
Implement a hash instead of mis-using a string pool.
---
ext/repo_rpmmd.c | 166 +++++++++++++++++++++++++++++++++++++++++--------------
1 file changed, 126 insertions(+), 40 deletions(-)
diff --git a/ext/repo_rpmmd.c b/ext/repo_rpmmd.c
index cc54b3e..77366c2 100644
--- a/ext/repo_rpmmd.c
+++ b/ext/repo_rpmmd.c
@@ -264,12 +264,12 @@ struct parsedata {
int first;
/* cspool ok to use */
int cshash_filled;
- /* Hash to maps checksums to solv */
- Stringpool cspool;
- /* Cache of known checksums to solvable id */
- Id *cshash;
- /* the current longest index in the table */
- int ncshash;
+
+ Hashtable cshash; /* checksum hash -> offset into csdata */
+ Hashval cshashm; /* hash mask */
+ int ncshash; /* entries used */
+ unsigned char *csdata; /* [len, checksum, id] */
+ int ncsdata; /* used bytes */
};
static Id
@@ -582,45 +582,125 @@ set_description_author(Repodata *data, Id handle, char *str, struct parsedata *p
}
+/*-----------------------------------------------*/
+/* checksum hash functions
+ *
+ * used to look up a solvable with the checksum for solvable extension purposes.
+ *
+ */
static void
init_cshash(struct parsedata *pd)
{
- /* initialize the string pool where we will store
- the package checksums we know about, to get an Id
- we can use in a cache */
- stringpool_init_empty(&pd->cspool);
}
static void
free_cshash(struct parsedata *pd)
{
- stringpool_free(&pd->cspool);
+ pd->cshash = solv_free(pd->cshash);
+ pd->ncshash = 0;
+ pd->cshashm = 0;
+ pd->csdata = solv_free(pd->csdata);
+ pd->ncsdata = 0;
+}
+
+static inline Hashval
+hashkey(const unsigned char *key, int keyl)
+{
+ return key[0] << 24 | key[1] << 16 | key[2] << 8 | key[3];
+}
+
+static void
+rebuild_cshash(struct parsedata *pd)
+{
+ Hashval h, hh, hm;
+ Hashtable ht;
+ unsigned char *d, *de;
+
+ hm = pd->cshashm;
+#if 0
+ fprintf(stderr, "rebuild cshash with mask 0x%x\n", hm);
+#endif
solv_free(pd->cshash);
+ ht = pd->cshash = (Hashtable)solv_calloc(hm + 1, sizeof(Id));
+ d = pd->csdata;
+ de = d + pd->ncsdata;
+ while (d != de)
+ {
+ h = hashkey(d + 1, d[0] + 1) & hm;
+ hh = HASHCHAIN_START;
+ while (ht[h])
+ h = HASHCHAIN_NEXT(h, hh, hm);
+ ht[h] = d + 1 - pd->csdata;
+ d += 2 + d[0] + sizeof(Id);
+ }
}
-/* save the checksum as key to solvable id relationship for
- metadata extension */
static void
-put_in_cshash(struct parsedata *pd, const char *key, Id id)
+put_in_cshash(struct parsedata *pd, const unsigned char *key, int keyl, Id id)
{
- Id index = stringpool_str2id(&pd->cspool, key, 1);
- if (index >= pd->ncshash)
+ Hashtable ht;
+ Hashval h, hh, hm;
+ unsigned char *d;
+
+ if (keyl < 4 || keyl > 256)
+ return;
+ ht = pd->cshash;
+ hm = pd->cshashm;
+ h = hashkey(key, keyl) & hm;
+ hh = HASHCHAIN_START;
+ if (ht)
{
- pd->cshash = solv_zextend(pd->cshash, pd->ncshash, index + 1 - pd->ncshash, sizeof(Id), 255);
- pd->ncshash = index + 1;
+ while (ht[h])
+ {
+ unsigned char *d = pd->csdata + ht[h];
+ if (d[-1] == keyl && !memcmp(key, d, keyl))
+ return; /* XXX: first id wins... */
+ h = HASHCHAIN_NEXT(h, hh, hm);
+ }
}
- /* add the checksum to the cache */
- pd->cshash[index] = id;
+ /* a new entry. put in csdata */
+ pd->csdata = solv_extend(pd->csdata, pd->ncsdata, 1, 1 + keyl + sizeof(Id), 4095);
+ d = pd->csdata + pd->ncsdata;
+ d[0] = keyl - 1;
+ memcpy(d + 1, key, keyl);
+ memcpy(d + 1 + keyl, &id, sizeof(Id));
+ pd->ncsdata += 1 + keyl + sizeof(Id);
+ if ((Hashval)++pd->ncshash * 2 > hm)
+ {
+ pd->cshashm = pd->cshashm ? (2 * pd->cshashm + 1) : 4095;
+ rebuild_cshash(pd);
+ }
+ else
+ ht[h] = pd->ncsdata - (keyl + sizeof(Id));
}
static Id
-lookup_cshash(struct parsedata *pd, const char *key)
+lookup_cshash(struct parsedata *pd, const unsigned char *key, int keyl)
{
- Id index = stringpool_str2id(&pd->cspool, key, 0);
- if (!index || index >= pd->ncshash || !pd->cshash[index])
+ Hashtable ht;
+ Hashval h, hh, hm;
+
+ if (keyl < 4 || keyl > 256)
+ return 0;
+ ht = pd->cshash;
+ if (!ht)
return 0;
- return pd->cshash[index];
+ hm = pd->cshashm;
+ h = hashkey(key, keyl) & hm;
+ hh = HASHCHAIN_START;
+ while (ht[h])
+ {
+ unsigned char *d = pd->csdata + ht[h];
+ if (d[-1] == keyl - 1 && !memcmp(key, d, keyl))
+ {
+ Id id;
+ memcpy(&id, d + keyl, sizeof(Id));
+ return id;
+ }
+ h = HASHCHAIN_NEXT(h, hh, hm);
+ }
+ return 0;
}
static void
@@ -630,14 +710,7 @@ fill_cshash_from_repo(struct parsedata *pd)
/* setup join data */
dataiterator_init(&di, pd->pool, pd->repo, 0, SOLVABLE_CHECKSUM, 0, 0);
while (dataiterator_step(&di))
- {
- const char *str;
-
- if (!solv_chksum_len(di.key->type))
- continue;
- str = repodata_chk2str(di.data, di.key->type, (const unsigned char *)di.kv.str);
- put_in_cshash(pd, str, di.solvid);
- }
+ put_in_cshash(pd, (const unsigned char *)di.kv.str, solv_chksum_len(di.key->type), di.solvid);
dataiterator_free(&di);
}
@@ -645,7 +718,7 @@ static void
fill_cshash_from_new_solvables(struct parsedata *pd)
{
Pool *pool = pd->pool;
- Id cstype;
+ Id cstype = 0;
unsigned const char *cs;
int i;
@@ -655,7 +728,7 @@ fill_cshash_from_new_solvables(struct parsedata *pd)
continue;
cs = repodata_lookup_bin_checksum_uninternalized(pd->data, i, SOLVABLE_CHECKSUM, &cstype);
if (cs)
- put_in_cshash(pd, repodata_chk2str(pd->data, cstype, cs), i);
+ put_in_cshash(pd, cs, solv_chksum_len(cstype), i);
}
}
@@ -750,13 +823,23 @@ startElement(void *userData, const char *name, const char **atts)
pd->extending = 0;
if ((pkgid = find_attr("pkgid", atts)) != NULL)
{
+ unsigned char chk[256];
+ int l;
+ const char *str = pkgid;
if (!pd->cshash_filled)
{
pd->cshash_filled = 1;
fill_cshash_from_new_solvables(pd);
}
+ handle = 0;
+ /* convert into bin checksum */
+ l = solv_hex2bin(&str, chk, sizeof(chk));
/* look at the checksum cache */
- handle = lookup_cshash(pd, pkgid);
+ if (l >= 4 && !pkgid[2 * l])
+ handle = lookup_cshash(pd, chk, l);
+#if 0
+ fprintf(stderr, "Lookup %s -> %d\n", pkgid, handle);
+#endif
if (!handle)
{
pool_debug(pool, SOLV_WARN, "the repository specifies extra information about package with checksum '%s', which does not exist in the repository.\n", pkgid);
@@ -1056,17 +1139,20 @@ endElement(void *userData, const char *name)
break;
case STATE_CHECKSUM:
{
- if (!pd->chksumtype)
+ unsigned char chk[256];
+ int l = solv_chksum_len(pd->chksumtype);
+ const char *str = pd->content;
+ if (!l || l > sizeof(chk))
break;
- if (strlen(pd->content) != 2 * solv_chksum_len(pd->chksumtype))
+ if (solv_hex2bin(&str, chk, l) != l || pd->content[2 * l])
{
- pd->ret = pool_error(pool, -1, "line %d: invalid checksum length for %s", (unsigned int)XML_GetCurrentLineNumber(*pd->parser), solv_chksum_type2str(pd->chksumtype));
+ pd->ret = pool_error(pool, -1, "line %u: invalid %s checksum", (unsigned int)XML_GetCurrentLineNumber(*pd->parser), solv_chksum_type2str(pd->chksumtype));
break;
}
- repodata_set_checksum(pd->data, handle, SOLVABLE_CHECKSUM, pd->chksumtype, pd->content);
+ repodata_set_bin_checksum(pd->data, handle, SOLVABLE_CHECKSUM, pd->chksumtype, chk);
/* we save the checksum to solvable id relationship for extending metadata */
if (pd->cshash_filled)
- put_in_cshash(pd, pd->content, s - pool->solvables);
+ put_in_cshash(pd, chk, l, s - pool->solvables);
break;
}
case STATE_FILE:
--
2.5.0